Dwi Kartini
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
Research Subjects
Dwi Kartini has a total of 73 co-authors. Here are the top co-authors by number of shared publications:
- Triando Hamonangan Saragih
Department of Computer Science, Lambung Mangkurat University, Banjarbaru, Indonesia
17 shared publications - Fatma Indriani
Department of Computer Science, Lambung Mangkurat University, Banjarbaru, Indonesia
14 shared publications - Mohammad Reza Faisal
Lambung Mangkurat University
13 shared publications - Irwan Budiman
Department of Computer Science, Lambung Mangkurat University, Banjarbaru, Indonesia
12 shared publications - Dodon Turianto Nugrahadi
Lambung Mangkurat University
9 shared publications - Muhammad Itqan Mazdadi
Lambung Mangkurat University
8 shared publications
Latest Publications
- Open Access
Characterising nuclear-norm matrix completion for nutrient imputation in food composition databases
- Open Access
- 1 June 2026
- Published by Poltekkes Kemenkes Surabaya in Indonesian Journal of electronics, electromedical engineering, and medical informatics
- Vol. 8 (3) , 365-379
Comparative Analysis of Image Augmentation and Class Weighting on ResNet50-CBAM for Pneumonia Detection from Chest X-Ray Images
Pneumonia remains one of the leading causes of morbidity and mortality worldwide, particularly among children, older adults, and immunocompromised individuals. Although chest X-ray (CXR) imaging is widely used for pneumonia diagnosis, manual interpretation is time-consuming, subjective, and highly dependent on radiologist expertise. Deep learning has shown promising performance for automated pneumonia classification; however, class imbalance remains a major challenge that can lead to biased predictions and reduced model generalization. Therefore, this study investigates the effectiveness of image augmentation and class weighting for handling class imbalance in pneumonia classification using chest X-ray images. The main contribution of this study is a systematic comparison of four experimental scenarios: Baseline, Augmentation Only, Class Weighting Only, and Hybrid (Image Augmentation and Class Weighting) implemented on a ResNet50 architecture integrated with the Convolutional Block Attention Module (CBAM). Experiments were conducted using the publicly available Chest X-Ray Images (Pneumonia) dataset from Kaggle, comprising 1,583 Normal and 4,273 Pneumonia images. Model performance was evaluated using Accuracy, ROC-AUC, Precision, Recall, F1-Score, confusion matrix analysis, and Youden’s J-Statistic for threshold optimization. The Hybrid model achieved the best overall performance, with an Accuracy of 95.74%, a ROC-AUC of 98.83%, a Macro Precision of 96.01%, a Macro Recall of 93.13%, and a Macro F1-Score of 94.44%. Moreover, the number of false negative predictions decreased from 25 in the Baseline model to 5 in the Hybrid model. These findings demonstrate that integrating image augmentation and class weighting within the ResNet50-CBAM framework effectively mitigates class imbalance and improves the reliability of automated pneumonia classification.
- 1 June 2026
- Published by Poltekkes Kemenkes Surabaya in Indonesian Journal of electronics, electromedical engineering, and medical informatics
- Vol. 8 (3) , 365-379
Open AccessComparative Analysis of Image Augmentation and Class Weighting on ResNet50-CBAM for Pneumonia Detection from Chest X-Ray Images
Pneumonia remains one of the leading causes of morbidity and mortality worldwide, particularly among children, older adults, and immunocompromised individuals. Although chest X-ray (CXR) imaging is widely used for pneumonia diagnosis, manual interpretation is time-consuming, subjective, and highly dependent on radiologist expertise. Deep learning has shown promising performance for automated pneumonia classification; however, class imbalance remains a major challenge that can lead to biased predictions and reduced model generalization. Therefore, this study investigates the effectiveness of image augmentation and class weighting for handling class imbalance in pneumonia classification using chest X-ray images. The main contribution of this study is a systematic comparison of four experimental scenarios: Baseline, Augmentation Only, Class Weighting Only, and Hybrid (Image Augmentation and Class Weighting) implemented on a ResNet50 architecture integrated with the Convolutional Block Attention Module (CBAM). Experiments were conducted using the publicly available Chest X-Ray Images (Pneumonia) dataset from Kaggle, comprising 1,583 Normal and 4,273 Pneumonia images. Model performance was evaluated using Accuracy, ROC-AUC, Precision, Recall, F1-Score, confusion matrix analysis, and Youden’s J-Statistic for threshold optimization. The Hybrid model achieved the best overall performance, with an Accuracy of 95.74%, a ROC-AUC of 98.83%, a Macro Precision of 96.01%, a Macro Recall of 93.13%, and a Macro F1-Score of 94.44%. Moreover, the number of false negative predictions decreased from 25 in the Baseline model to 5 in the Hybrid model. These findings demonstrate that integrating image augmentation and class weighting within the ResNet50-CBAM framework effectively mitigates class imbalance and improves the reliability of automated pneumonia classification.
- Open Access
- 1 June 2026
- Published by Institute of Advanced Engineering and Science in Bulletin of Electrical Engineering and Informatics
- Vol. 15 (3) , 2611-2624
Feature extraction and machine learning methods for biometric recognition based on fusion of ECG and fingerprint
This research introduces a multimodal biometric authentication framework by amalgamating electrocardiogram (ECG) and fingerprint modalities through the utilization of diverse feature extraction methodologies and machine learning classifiers. The proposed methodology aspires to augment precision and mitigate spoofing vulnerabilities in contrast to traditional single-modality systems. Among the feature extraction techniques assessed—grayscale, binary, Sobel edge detection, and minutiae—Naïve Bayes (NB) in conjunction with minutiae features exhibited superior performance, attaining an accuracy rate of 96.25%. Supplementary experiments employing random forest (RF) and support vector machine (SVM) also revealed commendable classification efficacy, underscoring the robustness of the fusion methodology. This investigation provides a pragmatic and secure biometric framework by harnessing complementary biometric characteristics to enhance authentication dependability. The proposed system presents promising applications in real-world contexts, particularly concerning mobile security and healthcare access control. Future research endeavors will tackle challenges associated with ECG signal variability, computational efficiency, and extensive deployment.
- 1 June 2026
- Published by Institute of Advanced Engineering and Science in Bulletin of Electrical Engineering and Informatics
- Vol. 15 (3) , 2611-2624
Open AccessFeature extraction and machine learning methods for biometric recognition based on fusion of ECG and fingerprint
This research introduces a multimodal biometric authentication framework by amalgamating electrocardiogram (ECG) and fingerprint modalities through the utilization of diverse feature extraction methodologies and machine learning classifiers. The proposed methodology aspires to augment precision and mitigate spoofing vulnerabilities in contrast to traditional single-modality systems. Among the feature extraction techniques assessed—grayscale, binary, Sobel edge detection, and minutiae—Naïve Bayes (NB) in conjunction with minutiae features exhibited superior performance, attaining an accuracy rate of 96.25%. Supplementary experiments employing random forest (RF) and support vector machine (SVM) also revealed commendable classification efficacy, underscoring the robustness of the fusion methodology. This investigation provides a pragmatic and secure biometric framework by harnessing complementary biometric characteristics to enhance authentication dependability. The proposed system presents promising applications in real-world contexts, particularly concerning mobile security and healthcare access control. Future research endeavors will tackle challenges associated with ECG signal variability, computational efficiency, and extensive deployment.
- Open Access
- 1 April 2026
- Published by Institute of Advanced Engineering and Science in Bulletin of Electrical Engineering and Informatics
- Vol. 15 (2) , 1417-1428
Detecting respiratory diseases using spectrogram-based deep features and machine learning algorithms
Early diagnosis of respiratory diseases is difficult as lung sound analysis requires the skills of medical professionals. Respiratory diseases are one of the leading causes of death in the world, so early detection is critical. Automatic identification is made possible by artificial intelligence. However, lung sound data is unstructured, while artificial intelligence often requires structured data. Therefore, feature extraction is required to structure the voice data. Traditional techniques such as mel-frequency cepstral coefficients (MFCC) often produce fewer features and information. This research uses a deep feature approach, which produces more features, as a solution. This research applies three convolutional neural network (CNN) architectures as deep features, namely VGG-16, DenseNet-121, and ResNet50, with machine learning classifications, namely random forest, support vector machine (SVM), Naïve Bayes, and K-nearest neighbors (KNN). This research will identify the optimal combination of methods. The results of this study show that respiratory disease classification can be effectively achieved by combining deep features and machine learning classification. The results of 10-fold cross-validation show that the three CNN architectures perform best on SVM with a linear kernel. The accuracy of VGG-16 is 70.63%, ResNet-50 is 64.93%, and DenseNet-121 is 73.58%.
- 1 April 2026
- Published by Institute of Advanced Engineering and Science in Bulletin of Electrical Engineering and Informatics
- Vol. 15 (2) , 1417-1428
Open AccessDetecting respiratory diseases using spectrogram-based deep features and machine learning algorithms
Early diagnosis of respiratory diseases is difficult as lung sound analysis requires the skills of medical professionals. Respiratory diseases are one of the leading causes of death in the world, so early detection is critical. Automatic identification is made possible by artificial intelligence. However, lung sound data is unstructured, while artificial intelligence often requires structured data. Therefore, feature extraction is required to structure the voice data. Traditional techniques such as mel-frequency cepstral coefficients (MFCC) often produce fewer features and information. This research uses a deep feature approach, which produces more features, as a solution. This research applies three convolutional neural network (CNN) architectures as deep features, namely VGG-16, DenseNet-121, and ResNet50, with machine learning classifications, namely random forest, support vector machine (SVM), Naïve Bayes, and K-nearest neighbors (KNN). This research will identify the optimal combination of methods. The results of this study show that respiratory disease classification can be effectively achieved by combining deep features and machine learning classification. The results of 10-fold cross-validation show that the three CNN architectures perform best on SVM with a linear kernel. The accuracy of VGG-16 is 70.63%, ResNet-50 is 64.93%, and DenseNet-121 is 73.58%.
- Open Access
- 26 November 2025
- Published by Poltekkes Kemenkes Surabaya in Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol. 8 (1) , 168-184
Deep Learning-Based Lung Sound Classification Using Mel-Spectrogram Features for Early Detection of Respiratory Diseases
Respiratory diseases such as asthma, chronic obstructive pulmonary disease, and pneumonia remain among the leading causes of death globally. Traditional diagnostic approaches, including auscultation, rely heavily on the subjective expertise of medical practitioners and the quality of the instruments used. Recent advancements in artificial intelligence offer promising alternatives for automated lung sound analysis. However, audio is an unstructured data format that must be converted into a suitable format for AI algorithms. Another significant challenge lies in the imbalanced class distribution within available datasets, which can adversely affect classification performance and model reliability. This study applied several comprehensive preprocessing techniques, including random undersampling to address data imbalance, resampling audio at 4000 Hz for standardization, and standardizing audio duration to 2.7 seconds for consistency. Feature extraction was then performed using the Mel Spectrogram method, converting audio signals into image representations to serve as input for classification algorithms based on deep learning architectures. To determine optimal performance characteristics, various Convolutional Neural Network (CNN) architectures were systematically evaluated, including LeNet-5, AlexNet, VGG-16, VGG-19, ResNet-50, and ResNet-152. VGG-16 achieved the highest classification accuracy of the tested models at 75.5%, demonstrating superior performance in respiratory sound classification tasks. This study demonstrates the potential of AI-based lung sound classification systems as a complementary diagnostic tool for healthcare professionals and the general public in supporting early identification of respiratory abnormalities and diseases. The findings suggest that automated lung sound analysis could enhance diagnostic accessibility and provide more valuable support for clinical decision-making in respiratory healthcare applications
- 26 November 2025
- Published by Poltekkes Kemenkes Surabaya in Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol. 8 (1) , 168-184
Open AccessDeep Learning-Based Lung Sound Classification Using Mel-Spectrogram Features for Early Detection of Respiratory Diseases
Respiratory diseases such as asthma, chronic obstructive pulmonary disease, and pneumonia remain among the leading causes of death globally. Traditional diagnostic approaches, including auscultation, rely heavily on the subjective expertise of medical practitioners and the quality of the instruments used. Recent advancements in artificial intelligence offer promising alternatives for automated lung sound analysis. However, audio is an unstructured data format that must be converted into a suitable format for AI algorithms. Another significant challenge lies in the imbalanced class distribution within available datasets, which can adversely affect classification performance and model reliability. This study applied several comprehensive preprocessing techniques, including random undersampling to address data imbalance, resampling audio at 4000 Hz for standardization, and standardizing audio duration to 2.7 seconds for consistency. Feature extraction was then performed using the Mel Spectrogram method, converting audio signals into image representations to serve as input for classification algorithms based on deep learning architectures. To determine optimal performance characteristics, various Convolutional Neural Network (CNN) architectures were systematically evaluated, including LeNet-5, AlexNet, VGG-16, VGG-19, ResNet-50, and ResNet-152. VGG-16 achieved the highest classification accuracy of the tested models at 75.5%, demonstrating superior performance in respiratory sound classification tasks. This study demonstrates the potential of AI-based lung sound classification systems as a complementary diagnostic tool for healthcare professionals and the general public in supporting early identification of respiratory abnormalities and diseases. The findings suggest that automated lung sound analysis could enhance diagnostic accessibility and provide more valuable support for clinical decision-making in respiratory healthcare applications
- Open Access
- 26 November 2025
- Published by Poltekkes Kemenkes Surabaya in Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol. 8 (1) , 270-283
Optimizing Input Window Length and Feature Requirements for Machine Learning-Based Postprandial Hyperglycemia Prediction
Continuous glucose monitoring systems currently generate alerts only after blood glucose thresholds are breached, limiting their utility for proactive diabetes management. Predicting postprandial glucose excursions before they occur requires determining the optimal amount of historical data and identifying which features contribute most to prediction accuracy. This study systematically evaluates how the length of the pre-meal observation window and feature composition affect machine-learning predictions of hyperglycemia events 60 minutes after eating. We analyzed 1,642 meal events from 45 adults wearing continuous glucose sensors, constructing features from pre-meal glucose trajectories, meal macronutrients, time of day, and health status. Four observation windows (15, 30, 45, 60 minutes) and three feature sets (all features, glucose-only, meal-only) were evaluated using Random Forest, XGBoost, and CatBoost with 5-fold group cross-validation. CatBoost with a 30-minute window achieved the best performance: 72.6% F1-macro, 79.6% accuracy, and 64.0% recall for hyperglycemia detection. Extending windows beyond 30 minutes did not yield consistent benefits, whereas 15-minute windows yielded comparable results. Glucose trajectory features alone retained 94% of full model performance (68.5% F1-macro), whereas meal composition alone proved insufficient (59.4% F1-macro). These findings demonstrate that recent glucose history dominates short-term prediction, enabling practical real-time systems with minimal data requirements. A 30-minute observation window with glucose and meal features offers an effective balance between prediction accuracy and system responsiveness.
- 26 November 2025
- Published by Poltekkes Kemenkes Surabaya in Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol. 8 (1) , 270-283
Open AccessOptimizing Input Window Length and Feature Requirements for Machine Learning-Based Postprandial Hyperglycemia Prediction
Continuous glucose monitoring systems currently generate alerts only after blood glucose thresholds are breached, limiting their utility for proactive diabetes management. Predicting postprandial glucose excursions before they occur requires determining the optimal amount of historical data and identifying which features contribute most to prediction accuracy. This study systematically evaluates how the length of the pre-meal observation window and feature composition affect machine-learning predictions of hyperglycemia events 60 minutes after eating. We analyzed 1,642 meal events from 45 adults wearing continuous glucose sensors, constructing features from pre-meal glucose trajectories, meal macronutrients, time of day, and health status. Four observation windows (15, 30, 45, 60 minutes) and three feature sets (all features, glucose-only, meal-only) were evaluated using Random Forest, XGBoost, and CatBoost with 5-fold group cross-validation. CatBoost with a 30-minute window achieved the best performance: 72.6% F1-macro, 79.6% accuracy, and 64.0% recall for hyperglycemia detection. Extending windows beyond 30 minutes did not yield consistent benefits, whereas 15-minute windows yielded comparable results. Glucose trajectory features alone retained 94% of full model performance (68.5% F1-macro), whereas meal composition alone proved insufficient (59.4% F1-macro). These findings demonstrate that recent glucose history dominates short-term prediction, enabling practical real-time systems with minimal data requirements. A 30-minute observation window with glucose and meal features offers an effective balance between prediction accuracy and system responsiveness.
- Open Access
- 26 November 2025
- Published by Poltekkes Kemenkes Surabaya in Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol. 8 (1) , 368-394
A Comparative Analysis of SMOTE and ADASYN for Cervical Cancer Detection using XGBoost with MICE Imputation
Cervical cancer remains a significant global health burden for women, with approximately 660,000 new cases and 350,000 associated deaths recorded worldwide in 2022. Machine learning methods have shown great promise in advancing timely detection and accurate diagnosis. This investigation compares two widely used oversampling strategies, Synthetic Minority Oversampling Technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN), applied to cervical cancer identification via the XGBoost classifier, paired with Multiple Imputation by Chained Equations (MICE) to handle incomplete data. The dataset consists of cervical cancer risk factors with four diagnostic outcomes: Hinselmann, Schiller, Cytology, and Biopsy, which are treated as independent binary classification tasks rather than a single multilabel classification problem. The process began by preparing a dataset of cervical cancer risk factors through MICE imputation, then applying SMOTE and ADASYN to address class imbalance. The XGBoost model is optimized using Random Search hyperparameter tuning and evaluated across train-test split ratios (50:50, 60:40, 70:30, 80:20, and 90:10) using accuracy, precision (macro, micro, weighted), recall (macro, micro, weighted), F1-score (macro, micro, weighted), and AUC metrics. The results indicated that the XGBoost setup with MICE and SMOTE outperformed the others, achieving 97.1% accuracy, 97.1% mic-precision, 97.1% mic-recall, 97.1% mic-F1, and 97.1% AUC. Meanwhile, the ADASYN-integrated model showed marginally lower results, with 95.4% accuracy, 95.4% micro-precision, 95.4% micro-recall, 95.4% micro-F1, and 55.5% AUC. SMOTE proved more adept at creating evenly distributed synthetic data for the underrepresented group. Overall, this work underscores the value of integrating MICE imputation, SMOTE oversampling, and tuned XGBoost as a reliable approach for cervical cancer detection. These insights pave the way for automated screening tools that can bolster clinical judgment and improve early diagnosis outcomes.
- 26 November 2025
- Published by Poltekkes Kemenkes Surabaya in Journal of Electronics, Electromedical Engineering, and Medical Informatics
- Vol. 8 (1) , 368-394
Open AccessA Comparative Analysis of SMOTE and ADASYN for Cervical Cancer Detection using XGBoost with MICE Imputation
Cervical cancer remains a significant global health burden for women, with approximately 660,000 new cases and 350,000 associated deaths recorded worldwide in 2022. Machine learning methods have shown great promise in advancing timely detection and accurate diagnosis. This investigation compares two widely used oversampling strategies, Synthetic Minority Oversampling Technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN), applied to cervical cancer identification via the XGBoost classifier, paired with Multiple Imputation by Chained Equations (MICE) to handle incomplete data. The dataset consists of cervical cancer risk factors with four diagnostic outcomes: Hinselmann, Schiller, Cytology, and Biopsy, which are treated as independent binary classification tasks rather than a single multilabel classification problem. The process began by preparing a dataset of cervical cancer risk factors through MICE imputation, then applying SMOTE and ADASYN to address class imbalance. The XGBoost model is optimized using Random Search hyperparameter tuning and evaluated across train-test split ratios (50:50, 60:40, 70:30, 80:20, and 90:10) using accuracy, precision (macro, micro, weighted), recall (macro, micro, weighted), F1-score (macro, micro, weighted), and AUC metrics. The results indicated that the XGBoost setup with MICE and SMOTE outperformed the others, achieving 97.1% accuracy, 97.1% mic-precision, 97.1% mic-recall, 97.1% mic-F1, and 97.1% AUC. Meanwhile, the ADASYN-integrated model showed marginally lower results, with 95.4% accuracy, 95.4% micro-precision, 95.4% micro-recall, 95.4% micro-F1, and 55.5% AUC. SMOTE proved more adept at creating evenly distributed synthetic data for the underrepresented group. Overall, this work underscores the value of integrating MICE imputation, SMOTE oversampling, and tuned XGBoost as a reliable approach for cervical cancer detection. These insights pave the way for automated screening tools that can bolster clinical judgment and improve early diagnosis outcomes.
- Open Access
- 30 August 2025
- Published by Universitas AMIKOM Purwokerto in Telematika
- Vol. 18 (2) , 91-104
Automatic Analysis of Natural Disaster Messages on Social Media Using IndoBERT and Multilingual BERT
As an important data source for mitigation processes and early warning systems. Social media platforms, such as X (formerly known as Twitter), have become primary channels for conveying real-time information, especially during disaster emergencies. With the large amount of unstructured disaster-related text that must be processed, the main challenge is accurately filtering and classifying messages into three categories: eyewitness, non-eyewitness, and don’t know. This research aims to compare the performance of four BERT-based natural language processing models, namely IndoBERT, IndoBERT with Masked Language Modeling (MLM), Multilingual BERT, and Multilingual BERT with MLM, in classifying Indonesian-language disaster messages. The dataset used in this study was obtained from previous research and publicly available data on GitHub, consisting of annotated messages related to floods, earthquakes, and forest fires. The method appl
- 30 August 2025
- Published by Universitas AMIKOM Purwokerto in Telematika
- Vol. 18 (2) , 91-104
Open AccessAutomatic Analysis of Natural Disaster Messages on Social Media Using IndoBERT and Multilingual BERT
As an important data source for mitigation processes and early warning systems. Social media platforms, such as X (formerly known as Twitter), have become primary channels for conveying real-time information, especially during disaster emergencies. With the large amount of unstructured disaster-related text that must be processed, the main challenge is accurately filtering and classifying messages into three categories: eyewitness, non-eyewitness, and don’t know. This research aims to compare the performance of four BERT-based natural language processing models, namely IndoBERT, IndoBERT with Masked Language Modeling (MLM), Multilingual BERT, and Multilingual BERT with MLM, in classifying Indonesian-language disaster messages. The dataset used in this study was obtained from previous research and publicly available data on GitHub, consisting of annotated messages related to floods, earthquakes, and forest fires. The method appl
- Open Access
- 20 August 2025
- Published by Poltekkes Kemenkes Surabaya in Indonesian Journal of electronics, electromedical engineering, and medical informatics
- Vol. 7 (3) , 539-548
Improving Diabetes Prediction Using Feedforward Neural Network with Adam Optimization and SMOTE Technique
Diabetes mellitus is a chronic metabolic disorder that demands early and accurate detection to prevent life-threatening complications. Traditional diagnostic procedures, such as blood glucose tests and oral glucose tolerance tests, are often invasive, time-consuming, and resource-intensive, making them less practical for widespread screening. This study aims to explore the potential of artificial intelligence, specifically Feedforward Neural Networks (FNN), in predicting diabetes based on clinical data from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). The main contribution of this research lies in the application of the Adaptive Moment Estimation (Adam) optimization algorithm and the Synthetic Minority Oversampling Technique (SMOTE) to enhance the performance and generalization of the FNN on imbalanced medical datasets. The methodology involves preprocessing steps such as imputing zero values with feature means, normalizing input features using Min-Max scaling, and applying SMOTE to balance class distribution. Two model configurations were compared: a baseline FNN trained manually using full-batch gradient descent and a second FNN optimized using Adam. Experimental results demonstrated that the baseline model achieved an accuracy of 70.13%, precision of 56.06%, recall of 68.52%, and F1-score of 61.67%, while the Adam-optimized model achieved superior results with an average accuracy of 73.31%, precision of 60.97%, recall of 66.67%, and F1-score of 63.64% across ten independent runs. These findings indicate that combining adaptive optimization with oversampling significantly enhances the robustness and reliability of neural networks for medical classification tasks. In conclusion, the proposed method provides an effective framework for AI-assisted early diabetes detection and opens pathways for future development using deeper network architectures and explainable AI models for clinical applications.
- 20 August 2025
- Published by Poltekkes Kemenkes Surabaya in Indonesian Journal of electronics, electromedical engineering, and medical informatics
- Vol. 7 (3) , 539-548
Open AccessImproving Diabetes Prediction Using Feedforward Neural Network with Adam Optimization and SMOTE Technique
Diabetes mellitus is a chronic metabolic disorder that demands early and accurate detection to prevent life-threatening complications. Traditional diagnostic procedures, such as blood glucose tests and oral glucose tolerance tests, are often invasive, time-consuming, and resource-intensive, making them less practical for widespread screening. This study aims to explore the potential of artificial intelligence, specifically Feedforward Neural Networks (FNN), in predicting diabetes based on clinical data from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). The main contribution of this research lies in the application of the Adaptive Moment Estimation (Adam) optimization algorithm and the Synthetic Minority Oversampling Technique (SMOTE) to enhance the performance and generalization of the FNN on imbalanced medical datasets. The methodology involves preprocessing steps such as imputing zero values with feature means, normalizing input features using Min-Max scaling, and applying SMOTE to balance class distribution. Two model configurations were compared: a baseline FNN trained manually using full-batch gradient descent and a second FNN optimized using Adam. Experimental results demonstrated that the baseline model achieved an accuracy of 70.13%, precision of 56.06%, recall of 68.52%, and F1-score of 61.67%, while the Adam-optimized model achieved superior results with an average accuracy of 73.31%, precision of 60.97%, recall of 66.67%, and F1-score of 63.64% across ten independent runs. These findings indicate that combining adaptive optimization with oversampling significantly enhances the robustness and reliability of neural networks for medical classification tasks. In conclusion, the proposed method provides an effective framework for AI-assisted early diabetes detection and opens pathways for future development using deeper network architectures and explainable AI models for clinical applications.
- Open Access
- 31 July 2025
- Published by Universitas Gadjah Mada in IJCCS (Indonesian Journal of Computing and Cybernetics Systems)
- Vol. 19 (3) , 327-336
Implementation of Chi-Square Feature Selection for Parkinson’s Disease Classification Using LightGBM
Penyakit Parkinson merupakan penyakit yang disebabkan oleh kerusakan sel saraf otak dan termasuk penyakit yang jumlah kasusnya meningkat pesat di dunia. Salah satu cara yang dapat dilakukan untuk mencegah meningkatnya kasus penyakit Parkinson adalah dengan melakukan diagnosis melalui metode klasifikasi dengan pendekatan pembelajaran algoritmik. Penelitian ini mengimplementasikan teknik Chi-Square untuk pendekatan pemilihan fitur yang relevan dengan algoritma Light Gradient Boosting Machine (LightGBM) dalam klasifikasi penyakit Parkinson. Pemilihan fitur Chi-Square bertujuan untuk mengurangi fitur yang kurang relevan sehingga dapat meningkatkan hasil kinerja model. Selain itu, metode SMOTE diterapkan untuk menangani ketidakseimbangan data dan penyetelan hiperparameter guna menentukan kombinasi parameter yang optimal. Pengujian dilakukan terhadap sepuluh variasi jumlah fitur, dengan hasil terbaik diperoleh dengan menggunakan 200 fitur yang menghasilkan akurasi sebesar 96,05%. Dengan menggunakan metode Chi-Square, kinerja model LightGBM meningkat dibandingkan dengan kinerja tanpa pemilihan fitur. Penerapan kombinasi metode ini dapat meningkatkan kinerja model klasifikasi secara signifikan dan berpotensi untuk diterapkan dalam sistem pendukung diagnosis penyakit Parkinson.
- 31 July 2025
- Published by Universitas Gadjah Mada in IJCCS (Indonesian Journal of Computing and Cybernetics Systems)
- Vol. 19 (3) , 327-336
Open AccessImplementation of Chi-Square Feature Selection for Parkinson’s Disease Classification Using LightGBM
Penyakit Parkinson merupakan penyakit yang disebabkan oleh kerusakan sel saraf otak dan termasuk penyakit yang jumlah kasusnya meningkat pesat di dunia. Salah satu cara yang dapat dilakukan untuk mencegah meningkatnya kasus penyakit Parkinson adalah dengan melakukan diagnosis melalui metode klasifikasi dengan pendekatan pembelajaran algoritmik. Penelitian ini mengimplementasikan teknik Chi-Square untuk pendekatan pemilihan fitur yang relevan dengan algoritma Light Gradient Boosting Machine (LightGBM) dalam klasifikasi penyakit Parkinson. Pemilihan fitur Chi-Square bertujuan untuk mengurangi fitur yang kurang relevan sehingga dapat meningkatkan hasil kinerja model. Selain itu, metode SMOTE diterapkan untuk menangani ketidakseimbangan data dan penyetelan hiperparameter guna menentukan kombinasi parameter yang optimal. Pengujian dilakukan terhadap sepuluh variasi jumlah fitur, dengan hasil terbaik diperoleh dengan menggunakan 200 fitur yang menghasilkan akurasi sebesar 96,05%. Dengan menggunakan metode Chi-Square, kinerja model LightGBM meningkat dibandingkan dengan kinerja tanpa pemilihan fitur. Penerapan kombinasi metode ini dapat meningkatkan kinerja model klasifikasi secara signifikan dan berpotensi untuk diterapkan dalam sistem pendukung diagnosis penyakit Parkinson.