Subhash Mondal

Artificial Intelligence has shown great promise in healthcare, particularly in non-invasive diagnostics using bio signals. This study focuses on classifying eye states (open or closed) using Electroencephalogram (EEG) signals captured via a 14-electrode neuroheadset, recorded through a Brain-Computer Interface (BCI). A publicly available dataset comprising 14,980 instances was used, where each sample represents EEG signals corresponding to eye activity. Fourteen classical machine learning (ML) models were evaluated using a tenfold cross-validation approach. The preprocessing pipeline involved removing outliers using the Z-score method, addressing class imbalance with SMOTETomek, and applying a bandpass filter to reduce signal noise. Significant EEG features were selected using a two-sample independent t-test (p < 0.05), ensuring only statistically relevant electrodes were retained. Additionally, the Common Spatial Pattern (CSP) method was used for feature extraction to enhance class separability by maximizing variance differences between eye states. Experimental results demonstrate that several classifiers achieved strong performance, with accuracy above 90%. The k-Nearest Neighbours classifier yielded the highest accuracy of 97.92% with CSP, and 97.75% without CSP. The application of CSP also enhanced the performance of Multi-Layer Perceptron and Support Vector Machine, reaching accuracies of 95.30% and 93.93%, respectively. The results affirm that integrating statistical validation, signal processing, and ML techniques can enable accurate and efficient EEG-based eye state classification, with practical implications for real-time BCI systems and offering a lightweight solution for real-time healthcare wearable applications healthcare applications.

Fetal health monitoring is critical for ensuring the well-being of both mother and fetus, with fetal electrocardiography (FECG) playing a pivotal role in detecting heart abnormalities such as arrhythmias and congenital defects. Traditional methods often struggle with challenges like noisy data and interference from maternal direct FECG, which hinder timely and accurate diagnoses. This research aims to develop an advanced deep learning-based classification system to detect abnormal FECG signals, monitor fetal heart patterns, and support healthcare providers with diagnostic insights for improved prenatal care. The study uses advanced deep learning models, such as LSTMs, GRUs, CNN-GRUs, and Transformers, to analyze time-series data from direct FECG recording signals. The dataset, sourced from the PhysioNet FECG database, underwent preprocessing approaches such as bandpass filtering, Independent Component Analysis (ICA), and R-peak detection, etc. to enhance signal quality. The novel CNN-GRU hybrid model achieved the highest performance, with an accuracy of 96.88% and an F1 score of 0.9836, outperforming standalone LSTM and GRU models with an accuracy of 93.75% and the Transformer model with an accuracy of 88.54%. The recommendation system added practical value by generating tailored reports for doctors and patients, bridging the gap between technical analysis and actionable healthcare guidance. By integrating accurate classification, advanced preprocessing, and actionable recommendations, this research study enhances early detection and monitoring of fetal heart conditions. This approach demonstrates the potential for AI-driven solutions to improve maternal-fetal health outcomes, paving the way for future real-time applications in prenatal care.

The growing use of computer resources in modern society makes it extremely vulnerable to several cyber-attacks, including unauthorized access to equipment and computer systems’ manipulation or utter breakdown. Malicious attacks in the form of malware cause significant harm to systems with limited resources. Hence, detecting these attacks and promptly implementing a computationally efficient technique is imperative. Utilizing a machine learning (ML) based model is a superior option for promptly identifying malware. This study develops fourteen machine learning models using a five-fold cross-validation technique on the dataset it obtained for research. We compute the execution time and memory used for each of the fourteen ML model developments, considering both all features and the reduced features after applying the data preprocessing methods. We utilized the Extra Tree classifier (ETC) to identify the top ten significant contributing features based on Gini impurity scores, which led to improved accuracy and reduced processing time. After that, we compared the experimental results and found that the Random Forest (RF) classification model on the reduced features set had a prediction accuracy of 99.39% and ROC-AUC values of 0.99. The ETC model prediction yields comparable results, confirming the viability of the suggested model. The proposed model is very resilient, exhibiting an extremely small standard deviation. It is also highly responsive, with reduced execution time and memory utilization.

Cardiovascular diseases (CVDs) continue to be a prominent cause of global mortality, necessitating the development of effective risk prediction models to combat the rise in heart disease (HD) mortality rates. This work presents a novel dual-stage stacked machine learning (ML) based computational risk prediction model for cardiac disorders. Leveraging a dataset that includes eleven significant characteristics from 1190 patients from five distinct sources, five ML classifiers are utilized to create the initial prediction model. To ensure robustness and generalizability, the classifiers are cross-validated ten times. The model performance is optimized by employing two hyperparameter tuning approaches: RandomizedSearchCV and GridSearchCV. These methods aim to find the optimal estimator values. The highest-performing models, specifically Random Forest, Extreme Gradient Boost, and Decision Tree undergo additional refinement using a stacking ensemble technique. The stacking model, which leverages the capabilities of the three models, attains a remarkable accuracy rate of 96%, a recall value of 0.98, and a ROC-AUC score of 0.96. Notably, the rate of false-negative results is below 1%, demonstrating a high level of accuracy and a non-overfitted model. To evaluate the model’s stability and repeatability, a comparable dataset consisting of 1000 occurrences is employed. The model consistently achieves an accuracy of 96.88% under identical experimental settings. This highlights the strength and dependability of the suggested computer model for predicting the risk of cardiac illnesses. The outcomes indicate that employing this two-step stacking ML method shows potential for prompt and precise diagnosis, hence aiding the worldwide endeavor to decrease fatalities caused by heart disease.

The Internet of Things (IoT) is vital as it offers extensive applicability in various fields, including healthcare. In the context of the risk level during pregnancy, to monitor and predict abnormalities, IoT devices provide a means to collect real-time health data, enabling continuous monitoring and analysis in the Internet of Medical Things (IoMT) environments. By integrating IoT devices into the system, crucial signs such as Heart Rate (HR), Systolic and Diastolic Blood Pressure (BP), Fetal Movements (FM), and Temperature (T) can be tracked remotely and non-invasively. This allows for the timely detection of abnormalities or potential risk factors during pregnancy, empowering healthcare professionals to intervene proactively and provide personalized care. This research focuses on developing a system for observing and predicting the maternal risk level in the IoT environment, mainly in remote areas. The goal is to improve maternal health and reduce maternal and child mortality rates, a significant decline according to United Nations targets for 2030. The research utilizes analytical tools and Machine Learning (ML) algorithms to analyze health data and risk factors associated with pregnancy. The acquired dataset contains various risk factors categorized and classified based on intensity. After comparing different ML models’ experimental results, Exploratory Data Analysis (EDA) approaches to determine the most effective risk factors. The fine-tuned Random Forest Classifier (RF) achieves the highest accuracy of 93.14%. An Android-based application has also been developed to deploy the prediction model to determine risk levels based on the different parameters.

Diabetic Retinopathy (DR) is a prevalent eye condition that occurs as a frequent complication among individuals with diabetes, particularly those who have been living with the disease for an extended period of time. This study uses fundus images to diagnose DR at five stages from early to late with No DR, Mild, Moderate, Severe, and Proliferative DR, commonly known as Stage 0 to Stage 4, respectively. This will aid in the timely treatment of diabetic patients preventing them from developing DR as early as possible. We used two most popular open-source datasets, the DR Detection database, namely APTOS 2019 and EyePACS, and combined them to create a larger dataset to trade off the data-centric obstacle and shortfall for any Deep Learning-based prediction models. Data augmentation and preprocessing techniques are applied to the images before feeding them to the proposed model to get a more accurate and efficient one. In the modern age oriented to Artificial Intelligence (AI), it is necessary to thoroughly analyze the identification of DR based on the existing Deep Learning (DL) models. After learning about the limitations of existing models, we have fine-tuned the ResNet50, DenseNet201 and InceptionV3 to enhance the model performance of the detection and categorization of DR. We have since proposed three Deep Convolutional Neural Networks (DCNN) models with better outcome based on accuracy than the existing state-of-the-art (SOTA) models. The fine-tuned DenseNet201 model, among the other two, performed significantly better with a validation accuracy of 90.04% and a negligible amount of loss, irrespective of each class, under the best configurable test conditions.

The thyroid gland plays a significant role in the human body's metabolism, growth, and development. Though it is not a life-threatening disease, a person suffering from thyroid faces many complications in their daily life. Recent trends have shown that women suffer more from thyroid-related diseases than men. The many contributing factors that lead to thyroid disease may be controlled upon early diagnosis stages. Machine learning prediction models help healthcare professionals diagnose thyroid diseases at an initial stage and take measures accordingly. This study deployed initial Sixteen ML models, including six boosting algorithms, on a dataset of 9172 instances with related features. The model performances have been judged through various standard performance metrics. The boosting algorithms showed exceptional results, and Cat Boost (CB) model produced the best accuracy of 95.75%. The hyperparameter tuning performed on boosting models by implementing Randomized Search CV increased the accuracy to 96.19% for CB. The stacking ensemble approach was applied on top of the six boosting tuned models with the CB classifier as the meta-learner. At the same time, the other boosting algorithms were kept as a base learner for the final model prediction. The accuracy of the stack model was impressive, with 95.32% compared with default models, the ROC-AUC at 0.95, and the other results were also promising. The model’s standard deviation was significantly less at 0.57, implying the model’s stability and robustness, and the False Negative (FN) rate reached 1.8%.