Philani Khumalo
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Philani Khumalo has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Rito Clifford Maswanganyi
Department of Computer and Electronic Engineering, Durban University of Technology, Durban 4001, South Africa
11 shared publications Bakhe Nleya
Department of Electronics and Computer Engineering
7 shared publications- Bakhe Nleya
Department of Electronic and Computer Engineering, Durban University of Technology, Steve Biko Campus, Durban 4001, South Africa
3 shared publications - Nerita Ramsoonder
Department of Computer and Electronic Engineering, Durban University of Technology, Durban 4001, South Africa
3 shared publications A. Mutsvangwa
Durban University of Technology,Electronic Eng Department,Durban,South Africa
2 shared publications- Andrew Mutsvangwa
Faculty of Engineering, Durban University of Technology, Durban, South Africa
2 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by MDPI AG in Algorithms
- Vol. 19 (11) , 978
Dataset Separability Versus Model Architecture in Electricity Theft Detection: A Rigorous Comparison of Four Machine Learning Models Across Two Datasets
Electricity theft leads to large amounts of non-technical losses for utility companies. Although the reported detection performance differs considerably, the various studies have used different algorithms, datasets and pipelines, so the variation cannot be assigned to any one factor. This study investigates what proportion of the difference is due to the architecture and what proportion is due to the dataset and the way in which it is labeled. The four methods, Random Forest, XGBoost, LightGBM and Isolation Forest, are compared using the same leakage-controlled pipeline with two datasets: the SGCC benchmark, which has labels derived from field inspections, and half-hourly consumption data from 53 schools in the Western Cape that has been subjected to injected suppression attacks. On the SGCC dataset, the supervised models achieve a theft class F1 score of 0.366 to 0.403 at accuracy levels of 0.88 to 0.90, with no significant pairwise differences (p = 0.126–0.867). With respect to the Western Cape dataset, XGBoost performs best (F1 0.727, FNR 0.294), once again without significance. Since there are only 17 positive cases, the smallest detectable difference in FNR is 0.450, and therefore these tests are uninformative about architecture rather than evidence of equivalence. Isolation Forest does worse on both datasets (ROC-AUC 0.640 and 0.565). None of the SGCC models outperforms the 91.47% majority baseline. When evaluated in the same score space, the binormal separability index increases from 1.203 to 2.060. The only F1 effect that is supported by the confidence intervals is the cross-dataset gap of +0.157 to +0.325. The reported differences are therefore due to the dataset and its labeling procedure, not to the architecture.
- 9 November 2026
- Published by MDPI AG in Algorithms
- Vol. 19 (11) , 978
Open AccessDataset Separability Versus Model Architecture in Electricity Theft Detection: A Rigorous Comparison of Four Machine Learning Models Across Two Datasets
Electricity theft leads to large amounts of non-technical losses for utility companies. Although the reported detection performance differs considerably, the various studies have used different algorithms, datasets and pipelines, so the variation cannot be assigned to any one factor. This study investigates what proportion of the difference is due to the architecture and what proportion is due to the dataset and the way in which it is labeled. The four methods, Random Forest, XGBoost, LightGBM and Isolation Forest, are compared using the same leakage-controlled pipeline with two datasets: the SGCC benchmark, which has labels derived from field inspections, and half-hourly consumption data from 53 schools in the Western Cape that has been subjected to injected suppression attacks. On the SGCC dataset, the supervised models achieve a theft class F1 score of 0.366 to 0.403 at accuracy levels of 0.88 to 0.90, with no significant pairwise differences (p = 0.126–0.867). With respect to the Western Cape dataset, XGBoost performs best (F1 0.727, FNR 0.294), once again without significance. Since there are only 17 positive cases, the smallest detectable difference in FNR is 0.450, and therefore these tests are uninformative about architecture rather than evidence of equivalence. Isolation Forest does worse on both datasets (ROC-AUC 0.640 and 0.565). None of the SGCC models outperforms the 91.47% majority baseline. When evaluated in the same score space, the binormal separability index increases from 1.203 to 2.060. The only F1 effect that is supported by the confidence intervals is the cross-dataset gap of +0.157 to +0.325. The reported differences are therefore due to the dataset and its labeling procedure, not to the architecture.
- Open Access
- 4 November 2026
- Published by MDPI AG in Big Data and Cognitive Computing
- Vol. 10 (11) , 374
Optimization of Signal Features in EEG-BCI Based on Multi-Class Subject-Specific Motor Imagery Classification
Motor Imagery (MI)-based Brain–Computer Interfaces (BCIs) often rely heavily on subject-specific feature selection to achieve high classification accuracy. However, this introduces substantial computational overhead and calibration friction that can limit clinical translation. This study benchmarks three distinct feature selection algorithms (Genetic Algorithm (GA), Differential Evolution (DE), Relief-F) to determine whether a universal feature set comprising common, shared features identified across subjects can be reliably established, thereby reducing or eliminating the need for subject-specific feature optimization. To evaluate this utilizing a within-subject approach, the methodology analyzes multi-class MI data sourced from the BCI Competition IV Dataset 2A and the PhysioNet EEGMI dataset. The processing pipeline applies Independent Component Analysis (ICA) for artifact removal, selecting a single clean representative component for each subject for analysis. Data augmentation is applied to increase trial size and improve classifier generalization. Features are extracted across multiple domains, consisting of time-domain metrics, Wavelet Packet Decomposition (WPD) nodes, band ratios, and absolute and relative spectral power. Classification is performed using K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), and a hybrid deep learning model leveraging the combined strengths of EEGNet and DeepConvNet. The evaluation reveals that DE-SVM achieves the best overall performance on the BCI Competition dataset with an average accuracy of 93.48%, while GA-SVM emerges as the top performer on PhysioNet with an accuracy of 89.71%. Furthermore, an ablation study confirms that SVM is the top-performing classifier across both datasets with average accuracies of 88.29% (BCI Competition) and 90.26% (PhysioNet). Statistical analysis of these ablation results confirms that a universal feature set can be robustly utilized within an MI-BCI system. The findings demonstrate that fixed, generalized feature subsets can effectively replace dynamic subject-specific optimization, offering a streamlined pathway toward more efficient and scalable BCI deployment.
- 4 November 2026
- Published by MDPI AG in Big Data and Cognitive Computing
- Vol. 10 (11) , 374
Open AccessOptimization of Signal Features in EEG-BCI Based on Multi-Class Subject-Specific Motor Imagery Classification
Motor Imagery (MI)-based Brain–Computer Interfaces (BCIs) often rely heavily on subject-specific feature selection to achieve high classification accuracy. However, this introduces substantial computational overhead and calibration friction that can limit clinical translation. This study benchmarks three distinct feature selection algorithms (Genetic Algorithm (GA), Differential Evolution (DE), Relief-F) to determine whether a universal feature set comprising common, shared features identified across subjects can be reliably established, thereby reducing or eliminating the need for subject-specific feature optimization. To evaluate this utilizing a within-subject approach, the methodology analyzes multi-class MI data sourced from the BCI Competition IV Dataset 2A and the PhysioNet EEGMI dataset. The processing pipeline applies Independent Component Analysis (ICA) for artifact removal, selecting a single clean representative component for each subject for analysis. Data augmentation is applied to increase trial size and improve classifier generalization. Features are extracted across multiple domains, consisting of time-domain metrics, Wavelet Packet Decomposition (WPD) nodes, band ratios, and absolute and relative spectral power. Classification is performed using K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), and a hybrid deep learning model leveraging the combined strengths of EEGNet and DeepConvNet. The evaluation reveals that DE-SVM achieves the best overall performance on the BCI Competition dataset with an average accuracy of 93.48%, while GA-SVM emerges as the top performer on PhysioNet with an accuracy of 89.71%. Furthermore, an ablation study confirms that SVM is the top-performing classifier across both datasets with average accuracies of 88.29% (BCI Competition) and 90.26% (PhysioNet). Statistical analysis of these ablation results confirms that a universal feature set can be robustly utilized within an MI-BCI system. The findings demonstrate that fixed, generalized feature subsets can effectively replace dynamic subject-specific optimization, offering a streamlined pathway toward more efficient and scalable BCI deployment.
- Open Access
- 22 October 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (11) , 344
Cross-Dataset Generalisation of Machine Learning and Deep Learning for Three-Class Alzheimer’s Disease Staging: A Six-Model Benchmark Including a Bidirectional State-Space Model
Alzheimer’s disease (AD) is the most common neurodegenerative disorder, and its pathology accumulates for years before symptoms appear, which places the mild cognitive impairment (MCI) stage at the centre of early detection. Machine learning (ML) and deep learning (DL) applied to structural magnetic resonance imaging (MRI) report high classification accuracy, yet these figures are usually measured within a single cohort and are rarely tested for transfer to an independent cohort, so whether they translate into clinical usefulness remains unresolved. The objective of this study is to address that gap directly. Six classifiers, three classical learners (Extra Trees, LightGBM and CatBoost) and three deep networks (ConvNeXt, CoAtNet and an introduced bidirectional Vision Mamba model, MambAD) were trained and tested on identical hippocampal slices from three publicly available datasets, namely ADNI, OASIS and a public Kaggle collection, for the three-class staging of cognitively normal (CN), MCI and AD. Every model was evaluated within cohorts and, more importantly, across cohorts, with subject-level scoring on the two clinical datasets. Within the primary ADNI cohort, the introduced model reached the highest balanced accuracy (BAC) at 0.578, although this lead was not statistically significant; across the six models, none dominated on every task, and no pairwise difference between models reached significance. BAC fell from about 0.55 within-cohort to between 0.38 and 0.45 across cohorts for every model, and the deep models transferred better on the ranking metric than the classical learners. The results show that a lightweight state-space model can match far larger pretrained backbones under matched conditions, and they confirm that single cohort accuracy overstates readiness, with cross-dataset generalisation remaining the binding constraint on clinical use.
- 22 October 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (11) , 344
Open AccessCross-Dataset Generalisation of Machine Learning and Deep Learning for Three-Class Alzheimer’s Disease Staging: A Six-Model Benchmark Including a Bidirectional State-Space Model
Alzheimer’s disease (AD) is the most common neurodegenerative disorder, and its pathology accumulates for years before symptoms appear, which places the mild cognitive impairment (MCI) stage at the centre of early detection. Machine learning (ML) and deep learning (DL) applied to structural magnetic resonance imaging (MRI) report high classification accuracy, yet these figures are usually measured within a single cohort and are rarely tested for transfer to an independent cohort, so whether they translate into clinical usefulness remains unresolved. The objective of this study is to address that gap directly. Six classifiers, three classical learners (Extra Trees, LightGBM and CatBoost) and three deep networks (ConvNeXt, CoAtNet and an introduced bidirectional Vision Mamba model, MambAD) were trained and tested on identical hippocampal slices from three publicly available datasets, namely ADNI, OASIS and a public Kaggle collection, for the three-class staging of cognitively normal (CN), MCI and AD. Every model was evaluated within cohorts and, more importantly, across cohorts, with subject-level scoring on the two clinical datasets. Within the primary ADNI cohort, the introduced model reached the highest balanced accuracy (BAC) at 0.578, although this lead was not statistically significant; across the six models, none dominated on every task, and no pairwise difference between models reached significance. BAC fell from about 0.55 within-cohort to between 0.38 and 0.45 across cohorts for every model, and the deep models transferred better on the ranking metric than the classical learners. The results show that a lightweight state-space model can match far larger pretrained backbones under matched conditions, and they confirm that single cohort accuracy overstates readiness, with cross-dataset generalisation remaining the binding constraint on clinical use.
- Open Access
- 20 October 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (10) , 338
AI-Based Electricity Theft Detection System: A Comparative Performance Analysis of Machine Learning and Deep Learning Models
Electricity theft is a worldwide challenge that causes significant losses for utility companies, with annual revenue losses estimated at USD 89.3 billion globally. The widespread rollout of Advanced Metering Infrastructure (AMI) and smart meters has made it easier to access large amounts of high-frequency consumption data from consumers. This access allows companies to use data-driven strategies for electricity theft detection. This review looks at how well machine learning (ML) and deep learning (DL) models detect electricity theft (ETD) using three benchmark datasets. The datasets presented are the State Grid Corporation of China (SGCC) dataset, the Open Energy Data Initiative Theft Detection Dataset 2022 (OEDI/TDD2022), and the Irish Commission for Energy Regulation Smart Metering Project (CER SMP) dataset. A thorough evaluation of 27 models is conducted, covering classical machine learning classifiers, tree ensemble methods, anomaly detection techniques, recurrent neural networks, convolutional networks, Transformer-based encoders, and federated learning systems. Key preprocessing methods, including filling in missing values, normalization, feature engineering, and tackling class imbalance with SMOTE and LoRAS, are noted for each model. The findings show that deep learning methods consistently outperform traditional machine learning techniques across all three benchmark datasets. The CNN-LSTM model, improved with LoRAS, achieves the top reported F1-score of 97.7% on the SGCC dataset. Meanwhile, the Federated ConvGRU model reaches the highest accuracy of 97.8% on OEDI/TDD2022. Transformer architectures show the best overall performance in average categories, with a mean accuracy of 95.8% and a mean F1-score of 95.5%. Addressing class imbalance is seen as the most influential factor, showing an 18.7 percentage-point difference in F1 scores between unaugmented and augmented LSTM models on SGCC. This review concludes that combining hybrid augmentation-driven deep learning architectures with privacy-preserving federated learning frameworks offers the most scalable and practical way to design Electricity theft detection (ETD) systems. The research community should prioritize standardized cross-dataset benchmarking and mandatory reporting of computational complexity as immediate priorities.
- 20 October 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (10) , 338
Open AccessAI-Based Electricity Theft Detection System: A Comparative Performance Analysis of Machine Learning and Deep Learning Models
Electricity theft is a worldwide challenge that causes significant losses for utility companies, with annual revenue losses estimated at USD 89.3 billion globally. The widespread rollout of Advanced Metering Infrastructure (AMI) and smart meters has made it easier to access large amounts of high-frequency consumption data from consumers. This access allows companies to use data-driven strategies for electricity theft detection. This review looks at how well machine learning (ML) and deep learning (DL) models detect electricity theft (ETD) using three benchmark datasets. The datasets presented are the State Grid Corporation of China (SGCC) dataset, the Open Energy Data Initiative Theft Detection Dataset 2022 (OEDI/TDD2022), and the Irish Commission for Energy Regulation Smart Metering Project (CER SMP) dataset. A thorough evaluation of 27 models is conducted, covering classical machine learning classifiers, tree ensemble methods, anomaly detection techniques, recurrent neural networks, convolutional networks, Transformer-based encoders, and federated learning systems. Key preprocessing methods, including filling in missing values, normalization, feature engineering, and tackling class imbalance with SMOTE and LoRAS, are noted for each model. The findings show that deep learning methods consistently outperform traditional machine learning techniques across all three benchmark datasets. The CNN-LSTM model, improved with LoRAS, achieves the top reported F1-score of 97.7% on the SGCC dataset. Meanwhile, the Federated ConvGRU model reaches the highest accuracy of 97.8% on OEDI/TDD2022. Transformer architectures show the best overall performance in average categories, with a mean accuracy of 95.8% and a mean F1-score of 95.5%. Addressing class imbalance is seen as the most influential factor, showing an 18.7 percentage-point difference in F1 scores between unaugmented and augmented LSTM models on SGCC. This review concludes that combining hybrid augmentation-driven deep learning architectures with privacy-preserving federated learning frameworks offers the most scalable and practical way to design Electricity theft detection (ETD) systems. The research community should prioritize standardized cross-dataset benchmarking and mandatory reporting of computational complexity as immediate priorities.
- Open Access
- 20 August 2026
- Published by MDPI AG in Big Data and Cognitive Computing
- Vol. 10 (8) , 280
Subject-Specific BCI Frameworks for Motor Imagery Classification Based on BSS-Free and BSS-Equipped Pipelines
The development of Motor Imagery (MI) Brain–Computer Interfaces (BCIs) is systematically constrained by low signal-to-noise ratios (SNRs), signal non-stationarity, and acute data scarcity. While complex Blind Source Separation (BSS) methods optimize signal clarity, their computational overhead introduces propagation delays that challenge real-time constraints. This study addresses this engineering trade-off by introducing a localized architectural framework to evaluate whether a lightweight pipeline operating without BSS (No-BSS) is sufficiently efficient for real-time control when compared against two BSS-equipped pipelines utilizing Independent Component Analysis (ICA) and Empirical Mode Decomposition (EMD). Validated across the BCI Competition IV Dataset 2A and the PhysioNet MI dataset, all three pipelines share an identical processing chain designed to maximize efficiency. To mitigate low SNRs, an Adaptive Laplacian spatial filter isolates neural intent across target sensorimotor electrodes (C3, C4, and Cz). Data scarcity is countered via a Gaussian noise injection data augmentation strategy, while session-to-session variability is addressed during feature extraction using Wavelet Packet Decomposition (WPD) paired with a Fisher Score criterion to dynamically isolate subject-specific time-frequency nodes. Redundant features are subsequently eliminated using a Genetic Algorithm (GA) before classification. Experimental evaluation reveals a distinct performance stratification: while the ICA (92.80%) and EMD (92.69%) pipelines yield the highest average accuracy for the PhysioNet dataset by isolating non-stationary and physiological noise, the No-BSS baseline (90.28%) remains the superior framework for the BCI Dataset 2A. Across all pipelines across both datasets, a stable classification hierarchy emerges wherein the Support Vector Machine (SVM) leads performance due to its maximum-margin decision boundary, followed by k-Nearest Neighbors (kNN), a modified EEGNet, and Decision Trees. The No-BSS baseline achieves classification accuracies highly competitive with its BSS counterparts while entirely bypassing their algorithmic overhead. Given the strict latency constraints of live BCI control loops, these findings establish the optimized No-BSS pipeline as a highly viable alternative for low-latency, real-time implementations.
- 20 August 2026
- Published by MDPI AG in Big Data and Cognitive Computing
- Vol. 10 (8) , 280
Open AccessSubject-Specific BCI Frameworks for Motor Imagery Classification Based on BSS-Free and BSS-Equipped Pipelines
The development of Motor Imagery (MI) Brain–Computer Interfaces (BCIs) is systematically constrained by low signal-to-noise ratios (SNRs), signal non-stationarity, and acute data scarcity. While complex Blind Source Separation (BSS) methods optimize signal clarity, their computational overhead introduces propagation delays that challenge real-time constraints. This study addresses this engineering trade-off by introducing a localized architectural framework to evaluate whether a lightweight pipeline operating without BSS (No-BSS) is sufficiently efficient for real-time control when compared against two BSS-equipped pipelines utilizing Independent Component Analysis (ICA) and Empirical Mode Decomposition (EMD). Validated across the BCI Competition IV Dataset 2A and the PhysioNet MI dataset, all three pipelines share an identical processing chain designed to maximize efficiency. To mitigate low SNRs, an Adaptive Laplacian spatial filter isolates neural intent across target sensorimotor electrodes (C3, C4, and Cz). Data scarcity is countered via a Gaussian noise injection data augmentation strategy, while session-to-session variability is addressed during feature extraction using Wavelet Packet Decomposition (WPD) paired with a Fisher Score criterion to dynamically isolate subject-specific time-frequency nodes. Redundant features are subsequently eliminated using a Genetic Algorithm (GA) before classification. Experimental evaluation reveals a distinct performance stratification: while the ICA (92.80%) and EMD (92.69%) pipelines yield the highest average accuracy for the PhysioNet dataset by isolating non-stationary and physiological noise, the No-BSS baseline (90.28%) remains the superior framework for the BCI Dataset 2A. Across all pipelines across both datasets, a stable classification hierarchy emerges wherein the Support Vector Machine (SVM) leads performance due to its maximum-margin decision boundary, followed by k-Nearest Neighbors (kNN), a modified EEGNet, and Decision Trees. The No-BSS baseline achieves classification accuracies highly competitive with its BSS counterparts while entirely bypassing their algorithmic overhead. Given the strict latency constraints of live BCI control loops, these findings establish the optimized No-BSS pipeline as a highly viable alternative for low-latency, real-time implementations.
- Open Access
- 27 July 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 140 (1) , 76
Design and Modeling of a Shunt Capacitor-Boosted Z-Source Inverter (SCB-ZSI)
Various inverter applications such as electronic vehicles, renewable energy systems, and uninterrupted power supplies have been the motive behind the increasing focus on the DC-AC inverters field. Therefore, DC-AC inverters have been evolving with the recent topology being the Z-source inverters (ZSIs). Though the ZSI overcame most of the limitations faced by the previous topologies such as the Voltage-Source Inverters (VSIs) and Current-Source Inverters (CSIs), they had shortcomings such as the increase in switching devices’ voltage stress and hence the deterioration of power quality with the increase in the boost factor. As a result, several ZSI-based topologies have been proposed in the literature to further improve the performance of a ZSI. This paper also proposes a different Z-source inverter topology called the Shunt Capacitor-Boosted Z-Source Inverter (SCB-ZSI) which seeks to improve the boost factor and lower the voltage stress. This inverter strategically adds two shunt capacitors on the impedance network of a traditional Z-source inverter, hence the Shunt Capacitor-Boosted-ZSI. The SCB-ZSI was mathematically modeled and simulated on MATLAB Simulink R2024a version. The SCB-ZSI was found to have a high boost factor compared to the ZSI for the same input DC voltage and modulation index. The SCB-ZSI was also found to incur less switching voltage stress across the switching devices compared to the ZSI for the same input DC voltage and modulation index. The simulation test results showed that the selection of shunt capacitors of a ZSI at 1% of those of the original ZSI improves the boost factor by 56% and reduces the switch voltage stress ration by 40% on an SCB-ZSI for the same set of input parameters. Though the SCB-ZSI is one of the promising ZSI-based inverter topologies, more work still has to be done before they can be industrially applied, such as developing an algorithm to design the shunt capacitors.
- 27 July 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 140 (1) , 76
Open AccessDesign and Modeling of a Shunt Capacitor-Boosted Z-Source Inverter (SCB-ZSI)
Various inverter applications such as electronic vehicles, renewable energy systems, and uninterrupted power supplies have been the motive behind the increasing focus on the DC-AC inverters field. Therefore, DC-AC inverters have been evolving with the recent topology being the Z-source inverters (ZSIs). Though the ZSI overcame most of the limitations faced by the previous topologies such as the Voltage-Source Inverters (VSIs) and Current-Source Inverters (CSIs), they had shortcomings such as the increase in switching devices’ voltage stress and hence the deterioration of power quality with the increase in the boost factor. As a result, several ZSI-based topologies have been proposed in the literature to further improve the performance of a ZSI. This paper also proposes a different Z-source inverter topology called the Shunt Capacitor-Boosted Z-Source Inverter (SCB-ZSI) which seeks to improve the boost factor and lower the voltage stress. This inverter strategically adds two shunt capacitors on the impedance network of a traditional Z-source inverter, hence the Shunt Capacitor-Boosted-ZSI. The SCB-ZSI was mathematically modeled and simulated on MATLAB Simulink R2024a version. The SCB-ZSI was found to have a high boost factor compared to the ZSI for the same input DC voltage and modulation index. The SCB-ZSI was also found to incur less switching voltage stress across the switching devices compared to the ZSI for the same input DC voltage and modulation index. The simulation test results showed that the selection of shunt capacitors of a ZSI at 1% of those of the original ZSI improves the boost factor by 56% and reduces the switch voltage stress ration by 40% on an SCB-ZSI for the same set of input parameters. Though the SCB-ZSI is one of the promising ZSI-based inverter topologies, more work still has to be done before they can be industrially applied, such as developing an algorithm to design the shunt capacitors.
- Open Access
- 22 July 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (7) , 217
Alzheimer’s Disease Detection Based on Machine Learning and Deep Learning Frameworks: A Cross-Dataset Comparative Performance Analysis and Assessment of Clinical Readiness
Alzheimer’s disease (AD) is the most prevalent neurodegenerative disorder worldwide, affecting approximately 56.9 million people in 2021 and projected to reach 152 million by 2050. Its defining pathological features, amyloid-beta plaques and neurofibrillary tangles, accumulate for up to two decades before cognitive symptoms emerge, placing the preclinical and mild cognitive impairment (MCI) stages at the centre of the early detection problem. Despite this, current diagnostic practice in routine clinical settings remains unreliable, with post-mortem studies placing the specificity of clinical AD diagnosis between 44.3 and 70.8% even in specialist memory clinics. Machine learning (ML) and deep learning (DL) applied to neuroimaging and electrophysiological data have emerged as candidate tools for closing this diagnostic gap, yet whether the accuracy figures reported in published studies translate into clinically useful performance on independent data remains unresolved. This study presents a structured comparative review of machine learning and deep learning methods reported across four publicly available Alzheimer’s disease datasets, namely the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the Open Access Series of Imaging Studies (OASIS), the OpenNeuro ds004504 electroencephalography (EEG) dataset, and the Kaggle Alzheimer’s magnetic resonance imaging (MRI) dataset. Thirteen model families are examined through the published literature rather than through new experiments, and for each model and dataset combination, the best accuracy reported in the source study is recorded alongside the model’s mathematical formulation. All performance figures reported in this abstract and throughout the paper are taken from the published studies reviewed, not from new experiments conducted by the authors. Across the reviewed studies, deep learning architectures pre-trained on ImageNet and fine-tuned on neuroimaging data are reported to produce the highest accuracy on MRI classification tasks. Residual Network (ResNet)-101 is reported at 98.21 percent on ADNI and 97.45 percent on OASIS, while the IncepRes fusion architecture reaches 98.35% on OASIS by combining multi-scale feature extraction from InceptionV3 with residual connectivity from ResNet152V2. Traditional machine learning classifiers remain competitive on tabular clinical and biomarker data, with Extreme Gradient Boosting (XGBoost) reaching 91% on ADNI multiclass features. Logistic Regression achieves 82 to 85% on binary ADNI classification and is the only classifier in this review that provides explicit per-feature prediction contributions without post hoc tooling. Gaussian Naïve Bayes achieves 80 to 83% on the same task. On the OpenNeuro EEG dataset, K-nearest neighbours (KNN) with singular value decomposition (SVD) entropy features achieves 91% binary accuracy, with feature engineering quality determining performance more reliably than classifier architecture. Eight principal findings emerge from the cross-dataset analysis. Binary classification consistently outperforms multiclass by 10 to 30% across all datasets, reflecting the genuine biological ambiguity of the mild cognitive impairment category. Dataset size and augmentation predict reported accuracy more reliably than model architecture. Ensemble methods outperform individual classifiers by 5 to 8% in nearly every imaging study. Deeper architectures can overfit small clinical cohorts. EEG models trail MRI models by approximately 10 to 15% on comparable binary tasks. Cross-dataset generalisation has not been systematically evaluated in most studies, and the few that have tested it report accuracy drops of 5 to 10% or more when models encounter data from different scanners or cohorts. Eight recurring limitations constrain the clinical utility of these findings. Small sample sizes and limited demographic diversity, severe class imbalance inflating raw accuracy metrics, poor cross-dataset generalisation driven by scanner heterogeneity, limited deep learning interpretability, the dominance of binary over multiclass tasks, the absence of longitudinal modelling despite available datasets, inadequate standardisation of preprocessing and evaluation protocols, and the signal-to-noise ratio constraints specific to EEG recordings of elderly patients collectively define the gap between benchmark performance and clinical readiness. Future work must prioritise multi-centre training cohorts, multimodal fusion architectures, longitudinal progression modelling, and standardised interpretability evaluation as non-optional requirements for any system intended for clinical deployment.
- 22 July 2026
- Published by MDPI AG in Machine Learning and Knowledge Extraction
- Vol. 8 (7) , 217
Open AccessAlzheimer’s Disease Detection Based on Machine Learning and Deep Learning Frameworks: A Cross-Dataset Comparative Performance Analysis and Assessment of Clinical Readiness
Alzheimer’s disease (AD) is the most prevalent neurodegenerative disorder worldwide, affecting approximately 56.9 million people in 2021 and projected to reach 152 million by 2050. Its defining pathological features, amyloid-beta plaques and neurofibrillary tangles, accumulate for up to two decades before cognitive symptoms emerge, placing the preclinical and mild cognitive impairment (MCI) stages at the centre of the early detection problem. Despite this, current diagnostic practice in routine clinical settings remains unreliable, with post-mortem studies placing the specificity of clinical AD diagnosis between 44.3 and 70.8% even in specialist memory clinics. Machine learning (ML) and deep learning (DL) applied to neuroimaging and electrophysiological data have emerged as candidate tools for closing this diagnostic gap, yet whether the accuracy figures reported in published studies translate into clinically useful performance on independent data remains unresolved. This study presents a structured comparative review of machine learning and deep learning methods reported across four publicly available Alzheimer’s disease datasets, namely the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the Open Access Series of Imaging Studies (OASIS), the OpenNeuro ds004504 electroencephalography (EEG) dataset, and the Kaggle Alzheimer’s magnetic resonance imaging (MRI) dataset. Thirteen model families are examined through the published literature rather than through new experiments, and for each model and dataset combination, the best accuracy reported in the source study is recorded alongside the model’s mathematical formulation. All performance figures reported in this abstract and throughout the paper are taken from the published studies reviewed, not from new experiments conducted by the authors. Across the reviewed studies, deep learning architectures pre-trained on ImageNet and fine-tuned on neuroimaging data are reported to produce the highest accuracy on MRI classification tasks. Residual Network (ResNet)-101 is reported at 98.21 percent on ADNI and 97.45 percent on OASIS, while the IncepRes fusion architecture reaches 98.35% on OASIS by combining multi-scale feature extraction from InceptionV3 with residual connectivity from ResNet152V2. Traditional machine learning classifiers remain competitive on tabular clinical and biomarker data, with Extreme Gradient Boosting (XGBoost) reaching 91% on ADNI multiclass features. Logistic Regression achieves 82 to 85% on binary ADNI classification and is the only classifier in this review that provides explicit per-feature prediction contributions without post hoc tooling. Gaussian Naïve Bayes achieves 80 to 83% on the same task. On the OpenNeuro EEG dataset, K-nearest neighbours (KNN) with singular value decomposition (SVD) entropy features achieves 91% binary accuracy, with feature engineering quality determining performance more reliably than classifier architecture. Eight principal findings emerge from the cross-dataset analysis. Binary classification consistently outperforms multiclass by 10 to 30% across all datasets, reflecting the genuine biological ambiguity of the mild cognitive impairment category. Dataset size and augmentation predict reported accuracy more reliably than model architecture. Ensemble methods outperform individual classifiers by 5 to 8% in nearly every imaging study. Deeper architectures can overfit small clinical cohorts. EEG models trail MRI models by approximately 10 to 15% on comparable binary tasks. Cross-dataset generalisation has not been systematically evaluated in most studies, and the few that have tested it report accuracy drops of 5 to 10% or more when models encounter data from different scanners or cohorts. Eight recurring limitations constrain the clinical utility of these findings. Small sample sizes and limited demographic diversity, severe class imbalance inflating raw accuracy metrics, poor cross-dataset generalisation driven by scanner heterogeneity, limited deep learning interpretability, the dominance of binary over multiclass tasks, the absence of longitudinal modelling despite available datasets, inadequate standardisation of preprocessing and evaluation protocols, and the signal-to-noise ratio constraints specific to EEG recordings of elderly patients collectively define the gap between benchmark performance and clinical readiness. Future work must prioritise multi-centre training cohorts, multimodal fusion architectures, longitudinal progression modelling, and standardised interpretability evaluation as non-optional requirements for any system intended for clinical deployment.
- Open Access
- 9 July 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 140 (1) , 75
A Utility-Driven Assessment of LoRaWAN Application for Secure Remote Monitoring in Smart Grid Systems
Low Power Wide Area Networks (LPWANs) are essential for enabling Internet-of-Things (IoT) technologies in utility environments. Utilities can leverage these networks to monitor critical remote assets, especially where mobile technologies are unsuitable due to poor power efficiency or insufficient coverage. This paper investigates the use of Long Range (LoRa) Wide Area Network (LoRaWAN) technology as an LPWAN solution for remote grid monitoring within the eThekwini Municipal Area. In addition to evaluating range performance (distance) and the packet reception ratio (PRR) across configurable parameters, such as spreading factor and transmit power, this paper introduces a data-packet security extension for LoRaWAN using NTRU post-quantum cryptography (PQC). The proposed security enhancement provides quantum-resistant encryption for application-layer payloads without violating LoRaWAN duty-cycle constraints or significantly increasing energy consumption. Field tests were performed at 11 geographically dispersed substations using a handheld LoRa device. Test signals were transmitted at four power levels (2 dBm, 8 dBm, 14 dBm, and 20 dBm) and spreading factors (SF7–SF12). Results show that the public LoRaWAN network can achieve communication distances of approximately 30 km in an urban environment, with PRR strongly dependent on SFs and transmit power. The integration of lightweight NTRU-protected payloads was found to be feasible for typical smart grid use cases involving small data packets (1–13 bytes).
- 9 July 2026
- Published by MDPI AG in Engineering Proceedings
- Vol. 140 (1) , 75
Open AccessA Utility-Driven Assessment of LoRaWAN Application for Secure Remote Monitoring in Smart Grid Systems
Low Power Wide Area Networks (LPWANs) are essential for enabling Internet-of-Things (IoT) technologies in utility environments. Utilities can leverage these networks to monitor critical remote assets, especially where mobile technologies are unsuitable due to poor power efficiency or insufficient coverage. This paper investigates the use of Long Range (LoRa) Wide Area Network (LoRaWAN) technology as an LPWAN solution for remote grid monitoring within the eThekwini Municipal Area. In addition to evaluating range performance (distance) and the packet reception ratio (PRR) across configurable parameters, such as spreading factor and transmit power, this paper introduces a data-packet security extension for LoRaWAN using NTRU post-quantum cryptography (PQC). The proposed security enhancement provides quantum-resistant encryption for application-layer payloads without violating LoRaWAN duty-cycle constraints or significantly increasing energy consumption. Field tests were performed at 11 geographically dispersed substations using a handheld LoRa device. Test signals were transmitted at four power levels (2 dBm, 8 dBm, 14 dBm, and 20 dBm) and spreading factors (SF7–SF12). Results show that the public LoRaWAN network can achieve communication distances of approximately 30 km in an urban environment, with PRR strongly dependent on SFs and transmit power. The integration of lightweight NTRU-protected payloads was found to be feasible for typical smart grid use cases involving small data packets (1–13 bytes).
- Open Access
- 30 April 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (9) , 1527
Subject-Specific Comparative Performance Analysis of Deep Learning Architectures for Motor Imagery Classification
Motor Imagery (MI)-based brain–computer interfaces (BCIs) offer promising solutions for enhancing communication and motor functions in individuals with neurological impairments. However, decoding EEG signals accurately is difficult because of their poor signal-to-noise ratio and variability across subjects and sessions. In addition, EEG signals are sensitive to noise. Moreover, the low spatial resolution of EEG signals makes model generalization unreliable due to differences between signals across subjects. While several deep learning models have been developed, a fair comparison remains difficult due to differences in pre-processing, training procedures, and evaluation protocols. This study provides a systematic, controlled comparison of five deep learning approaches for subject-specific classification—EEGNet, EEG-TCNet, ShallowConvNet, DeepConvNet, and CTNet—using the BCI Competition IV datasets 2a and 2b. To enable an unbiased comparison, all models are trained using the same pipeline, with uniform pre-processing and training. Apart from classical accuracy scores, the effect of a constant set of hyper-parameters on the training dynamics, generalization capacity, and the susceptibility to overfitting is evaluated. The performance of the above-stated models is evaluated based on training dynamics, computational efficiency, accuracy, and the quality of the features learned by the models. Using the five-dimensional analysis framework consisting of quantitative performance metrics, training curves, confusion matrix analysis, ROC analysis, and t-SNE visualization techniques, the performance of the brain–computer interfaces is comprehensively analyzed. The experimental analysis confirms that CTNet outperforms other models, with accuracy values of 82.56% and 86.42% on the BCI competition IV datasets 2a and 2b, respectively. The EEGNet model is recognized as having the most potential in the field of real-time applications, owing to its light structure; meanwhile, the DeepConvNet model shows signs of overfitting, despite showing good accuracy. These findings highlight that model training characteristics and sensitivity to the hyper-parameters are important factors in evaluating deep learning models for MI-EEG classification problems.
- 30 April 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (9) , 1527
Open AccessSubject-Specific Comparative Performance Analysis of Deep Learning Architectures for Motor Imagery Classification
Motor Imagery (MI)-based brain–computer interfaces (BCIs) offer promising solutions for enhancing communication and motor functions in individuals with neurological impairments. However, decoding EEG signals accurately is difficult because of their poor signal-to-noise ratio and variability across subjects and sessions. In addition, EEG signals are sensitive to noise. Moreover, the low spatial resolution of EEG signals makes model generalization unreliable due to differences between signals across subjects. While several deep learning models have been developed, a fair comparison remains difficult due to differences in pre-processing, training procedures, and evaluation protocols. This study provides a systematic, controlled comparison of five deep learning approaches for subject-specific classification—EEGNet, EEG-TCNet, ShallowConvNet, DeepConvNet, and CTNet—using the BCI Competition IV datasets 2a and 2b. To enable an unbiased comparison, all models are trained using the same pipeline, with uniform pre-processing and training. Apart from classical accuracy scores, the effect of a constant set of hyper-parameters on the training dynamics, generalization capacity, and the susceptibility to overfitting is evaluated. The performance of the above-stated models is evaluated based on training dynamics, computational efficiency, accuracy, and the quality of the features learned by the models. Using the five-dimensional analysis framework consisting of quantitative performance metrics, training curves, confusion matrix analysis, ROC analysis, and t-SNE visualization techniques, the performance of the brain–computer interfaces is comprehensively analyzed. The experimental analysis confirms that CTNet outperforms other models, with accuracy values of 82.56% and 86.42% on the BCI competition IV datasets 2a and 2b, respectively. The EEGNet model is recognized as having the most potential in the field of real-time applications, owing to its light structure; meanwhile, the DeepConvNet model shows signs of overfitting, despite showing good accuracy. These findings highlight that model training characteristics and sensitivity to the hyper-parameters are important factors in evaluating deep learning models for MI-EEG classification problems.
- Open Access
- 30 April 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (9) , 1520
Comparative Performance Analysis of Machine Learning Computational Pipelines and Deep Learning Architectures in EEG Motor Imagery BCIs
The deployment of Motor Imagery Brain–Computer Interfaces (MI-BCI) is constrained by the inherent physiological variabilities of Electroencephalography (EEG) and parametric opacity. This paper presents a targeted technical audit of ten high-density MI-BCI computational pipelines, evaluating how existing literature addresses low Signal-to-Noise Ratio (SNR), intra-subject variability, and session-to-session instability. The investigation focuses on the contamination of data by ocular and muscular artifacts that overlap with the spectral components of Mu and Beta rhythms, often leading to algorithmic overfitting. Furthermore, the paper evaluates the impact of manifold drift where fluctuations in user state necessitate frequent recalibration as a primary hurdle for BCI portability. By applying a forensic evaluation framework to standardize the analysis across the ten selected studies, this paper identifies a high-performance landscape within standardized benchmarks, with classification accuracies reaching peak values of 95.42%. The audit specifically identifies a performance-reporting gap; while hybrid architectures demonstrate superior noise-rejection, they are frequently characterized by undocumented computational overhead. Additionally, while Neighborhood Component Analysis (NCA) emerges as a stable feature selection algorithm across the sampled literature, the systemic absence of reported execution times prevents a verified assessment of its low-latency viability. A critical technical finding is the widespread issue of Parametric Opacity, particularly regarding the omission of essential deterministic variables such as filter orders, windowing constants, and the final dimensionality of feature vectors. The audit reveals that the frequent failure to report the exact number of features utilized for classification masks potential overfitting and prevents an accurate assessment of the system’s generalization capabilities. Furthermore, only a specialized subset of the reviewed literature validates performance through formal statistical testing, such as Friedman ANOVA or Wilcoxon Signed-Rank tests, with most studies relying on peak accuracy metrics that may disguise filtered artifact residuals. This lack of granular documentation disguises the computational complexity of proposed methods and complicates their feasibility for hardware-in-the-loop validation. The findings establish that standardizing the reporting of preprocessing variables and feature-space dimensions is a prerequisite for overcoming current performance plateaus in universal BCI architectures.
- 30 April 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (9) , 1520
Open AccessComparative Performance Analysis of Machine Learning Computational Pipelines and Deep Learning Architectures in EEG Motor Imagery BCIs
The deployment of Motor Imagery Brain–Computer Interfaces (MI-BCI) is constrained by the inherent physiological variabilities of Electroencephalography (EEG) and parametric opacity. This paper presents a targeted technical audit of ten high-density MI-BCI computational pipelines, evaluating how existing literature addresses low Signal-to-Noise Ratio (SNR), intra-subject variability, and session-to-session instability. The investigation focuses on the contamination of data by ocular and muscular artifacts that overlap with the spectral components of Mu and Beta rhythms, often leading to algorithmic overfitting. Furthermore, the paper evaluates the impact of manifold drift where fluctuations in user state necessitate frequent recalibration as a primary hurdle for BCI portability. By applying a forensic evaluation framework to standardize the analysis across the ten selected studies, this paper identifies a high-performance landscape within standardized benchmarks, with classification accuracies reaching peak values of 95.42%. The audit specifically identifies a performance-reporting gap; while hybrid architectures demonstrate superior noise-rejection, they are frequently characterized by undocumented computational overhead. Additionally, while Neighborhood Component Analysis (NCA) emerges as a stable feature selection algorithm across the sampled literature, the systemic absence of reported execution times prevents a verified assessment of its low-latency viability. A critical technical finding is the widespread issue of Parametric Opacity, particularly regarding the omission of essential deterministic variables such as filter orders, windowing constants, and the final dimensionality of feature vectors. The audit reveals that the frequent failure to report the exact number of features utilized for classification masks potential overfitting and prevents an accurate assessment of the system’s generalization capabilities. Furthermore, only a specialized subset of the reviewed literature validates performance through formal statistical testing, such as Friedman ANOVA or Wilcoxon Signed-Rank tests, with most studies relying on peak accuracy metrics that may disguise filtered artifact residuals. This lack of granular documentation disguises the computational complexity of proposed methods and complicates their feasibility for hardware-in-the-loop validation. The findings establish that standardizing the reporting of preprocessing variables and feature-space dimensions is a prerequisite for overcoming current performance plateaus in universal BCI architectures.