Tanmoy Sarkar Pias

All Publications

• Optimizing stability of heart disease prediction across imbalanced learning with interpretable Grow Network COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE Bin Akter, S., Akter, S., Hasan, R., Hasan, M., Eisenberg, D., Azim, R., Fernandez, J., Pias, T. 2025; 265: 108702

  Abstract

  Heart disease prediction models often face stability challenges when applied to public datasets due to significant class imbalances, unlike the more balanced benchmark datasets. These imbalances can adversely affect various stages of prediction, including feature selection, sampling, and modeling, leading to skewed performance, with one class often being favored over another.To enhance stability, this study proposes a Grow Network (GrowNet) architecture, which dynamically configures itself based on the data's characteristics. To enhance GrowNet's stability, this study proposes the use of TriDyn Dependence feature selection and Adaptive Refinement sampling, which ensure the selection of relevant features across imbalanced data and effectively manage class imbalance during training.When evaluated on the benchmark UCI heart disease dataset, GrowNet has outperformed other models, achieving a specificity of 92%, sensitivity of 88%, precision of 90%, and F1 score of 90%. Further evaluation on three public datasets from the Behavioral Risk Factor Surveillance System (BRFSS), where heart disease cases constitute only about 6% of the data, has demonstrated GrowNet's ability to maintain balanced performance, with an average specificity, sensitivity, and AUC-ROC of 77.67%, 81.67%, and 89.67%, respectively, while other models have exhibited instability. This represents a 22.8% improvement in handling class imbalance compared to prior studies. Additional tests on two public datasets from the National Health Interview Survey (NHIS) have confirmed GrowNet's robustness and generalizability, with an average specificity, sensitivity, and AUC-ROC of 80.5%, 82.5%, and 90%, respectively, while other models have continued to demonstrate instability.To enhance transparency, this study incorporates SHapley Additive exPlanations (SHAP) analysis, enabling healthcare professionals to understand the decision-making process and identify key risk factors for heart disease, such as bronchitis in midlife, renal dysfunction in the elderly, and depressive disorders in individuals aged 35-44.This study presents a robust, interpretable model to assist healthcare professionals in cost-effective, early heart disease detection by focusing on key risk factors, ultimately improving patient outcomes.

  View details for DOI 10.1016/j.cmpb.2025.108702

  View details for Web of Science ID 001458919000001

  View details for PubMedID 40147157

• Low responsiveness of machine learning models to critical or deteriorating health conditions COMMUNICATIONS MEDICINE Pias, T., Afrose, S., Das Tuli, M., Trisha, I., Deng, X., Nemeroff, C. B., Yao, D. 2025; 5 (1): 62

  Abstract

  Machine learning (ML) based mortality prediction models can be immensely useful in intensive care units. Such a model should generate warnings to alert physicians when a patient's condition rapidly deteriorates, or their vitals are in highly abnormal ranges. Before clinical deployment, it is important to comprehensively assess a model's ability to recognize critical patient conditions.We develop multiple medical ML testing approaches, including a gradient ascent method and neural activation map. We systematically assess these machine learning models' ability to respond to serious medical conditions using additional test cases, some of which are time series. Guided by medical doctors, our evaluation involves multiple machine learning models, resampling techniques, and four datasets for two clinical prediction tasks.We identify serious deficiencies in the models' responsiveness, with the models being unable to recognize severely impaired medical conditions or rapidly deteriorating health. For in-hospital mortality prediction, the models tested using our synthesized cases fail to recognize 66% of the injuries. In some instances, the models fail to generate adequate mortality risk scores for all test cases. Our study identifies similar kinds of deficiencies in the responsiveness of 5-year breast and lung cancer prediction models.Using generated test cases, we find that statistical machine-learning models trained solely from patient data are grossly insufficient and have many dangerous blind spots. Most of the ML models tested fail to respond adequately to critically ill patients. How to incorporate medical knowledge into clinical machine learning models is an important future research direction.

  View details for DOI 10.1038/s43856-025-00775-0

  View details for Web of Science ID 001442150700003

  View details for PubMedID 40069422

  View details for PubMedCentralID PMC11897252

• Early detection of subjective cognitive decline from self-reported symptoms: An interpretable attention-cost fusion approach JOURNAL OF BIOMEDICAL INFORMATICS Bin Akter, S., Akter, S., Hasan, R., Hasan, M., Islam, A., Pias, T., Fernandez, J., Alam, M., Eisenberg, D. 2025; 162: 104770

  Abstract

  Subjective cognitive decline (SCD) refers to self-reported difficulties in concentration, memory, and decision-making. Since SCD is based on subjective experiences, no specific medical test can definitively confirm its presence, making early detection challenging. Thus, it is advantageous to develop an AI model to capitalize on self-reported health complications, personality traits, and sociodemographic factors for early detection of SCD.This research has proposed an AI-based framework for SCD detection using self-reported measures from the BRFSS 2021 dataset. A novel Weighted Fusion Selection (WFS) approach has been introduced, which combines multiple feature selection techniques to address class imbalance and identify relevant features associated with less frequent classes. The data set has shown a significant imbalance, with individuals at risk of SCD being 81.76% fewer than those not at risk. An Attention Cost Convolutional Neural Network (AC-CNN) has been developed to address this, integrating channel-wise attention mechanisms and cost-sensitive learning to enhance performance across imbalanced data.The AC-CNN model has achieved a balance between specificity (77%) and sensitivity (81%), with an AUC-ROC of 0.87. This has represented an overall 24.8% improvement in handling class imbalance compared to prior studies. Additional testing on the NHIS 2022 dataset has shown that AC-CNN has maintained balanced performance, confirming its robust generalizability, while other models have remained unstable.Further, applying SHapley Additive exPlanations (SHAP) explainable techniques to the AC-CNN model has revealed how individual aspects of an individual's health records, lifestyle, and demographics contribute to the prediction of SCD. For example, depression, low education, poor income, inadequate healthcare, and poor overall health have all been strongly linked to an increased risk of SCD.

  View details for DOI 10.1016/j.jbi.2024.104770

  View details for Web of Science ID 001397447300001

  View details for PubMedID 39756527

• Fair and explainable Myocardial Infarction (MI) prediction: Novel strategies for feature selection and class imbalance correction. Computers in biology and medicine Akter, S. B., Akter, S., Tuli, M. D., Eisenberg, D., Lotvola, A., Islam, H., Fernandez, J. F., Huttemann, M., Pias, T. S. 2025; 184: 109413

  Abstract

  The rising incidences of myocardial infarction (MI), often affecting individuals without traditional risk factors, highlight the urgent need for improved early detection using personal health data. However, health surveys and electronic health records (EHRs) frequently suffer from class imbalances, leading to prediction biases and differences between specificity and sensitivity, which hinder reliable model development despite the valuable insights contained in these datasets. To address this, we have introduced a novel approach to enhance MI risk prediction using self-reported attributes from the Behavioral Risk Factor Surveillance System (BRFSS) and the National Health Interview Survey (NHIS) dataset. Our approach incorporates three innovative techniques: the Dual-Path Artificial Neural Network (DP-ANN) to mitigate biased decision making across imbalanced datasets, the Triple Criteria Selection (TCS) for unbiased feature selection, and Minority Weighted Sampling (MWS) to tackle challenges of uncontrolled minority class sampling. These methods collectively enhance MI prediction and feature relevance. The DP-ANN model has achieved balanced performance, with an average specificity of 80%, sensitivity of 82%, and AUC-ROC of 89.5%, improving imbalance variance by approximately 14.96% compared to prior studies. By outperforming other models across four heavily imbalanced datasets, our approach demonstrates robustness and generalizability. Additionally, SHapley Additive exPlanations (SHAP) analysis has revealed key predictors and risk factors for MI, such as coronary heart disease and bronchitis in females, and stroke among individuals aged 35-54. In conclusion, our study provides a robust model for healthcare professionals to assess MI risk through targeted factors, promoting early detection and potentially improving patient outcomes.

  View details for DOI 10.1016/j.compbiomed.2024.109413

  View details for PubMedID 39615231

• Methods and Benchmark for Detecting Cryptographic API Misuses in Python IEEE TRANSACTIONS ON SOFTWARE ENGINEERING Frantz, M., Xiao, Y., Pias, T., Meng, N., Yao, D. 2024; 50 (5): 1118-1129

  View details for DOI 10.1109/TSE.2024.3377182

  View details for Web of Science ID 001224187300014

• Ensemble learning based transmission line fault classification using phasor measurement unit (PMU) data with explainable AI (XAI) PLOS ONE Bin Akter, S., Pias, T., Deeba, S., Hossain, J., Rahman, H. 2024; 19 (2): e0295144

  Abstract

  A large volume of data is being captured through the Phasor Measurement Unit (PMU), which opens new opportunities and challenges to the study of transmission line faults. To be specific, the Phasor Measurement Unit (PMU) data represents many different states of the power networks. The states of the PMU device help to identify different types of transmission line faults. For a precise understanding of transmission line faults, only the parameters that contain voltage and current magnitude estimations are not sufficient. This requirement has been addressed by generating data with more parameters such as frequencies and phase angles utilizing the Phasor Measurement Unit (PMU) for data acquisition. The data has been generated through the simulation of a transmission line model on ePMU DSA tools and Matlab Simulink. Different machine learning models have been trained with the generated synthetic data to classify transmission line fault cases. The individual models including Decision Tree (DT), Random Forest (RF), and K-Nearest Neighbor (K-NN) have outperformed other models in fault classification which have acquired a cross-validation accuracy of 99.84%, 99.83%, and 99.76% respectively across 10 folds. Soft voting has been used to combine the performance of these best-performing models. Accordingly, the constructed ensemble model has acquired a cross-validation accuracy of 99.88% across 10 folds. The performance of the combined models in the ensemble learning process has been analyzed through explainable AI (XAI) which increases the interpretability of the input parameters in terms of making predictions. Consequently, the developed model has been evaluated with several performance matrices, such as precision, recall, and f1 score, and also tested on the IEEE 14 bus system. To sum up, this article has demonstrated the classification of six scenarios including no fault and fault cases from transmission lines with a significant number of training parameters and also interpreted the effect of each parameter to make predictions of different fault cases with great success.

  View details for DOI 10.1371/journal.pone.0295144

  View details for Web of Science ID 001163966900041

  View details for PubMedID 38346050

  View details for PubMedCentralID PMC10861062

• Neuromarketing Techniques to Enhance Consumer Preference Prediction Eisenberg, D., Pias, T., Fjermestad, J., Fresneda, J. edited by Bui, T. X. HICSS. 2024: 923-932

  View details for Web of Science ID 001301787501002

• M1M2: Deep-Learning-Based Real-Time Emotion Recognition from Neural Activity SENSORS Akter, S., Prodhan, R., Pias, T., Eisenberg, D., Fernandez, J. 2022; 22 (21)

  Abstract

  Emotion recognition, or the ability of computers to interpret people's emotional states, is a very active research area with vast applications to improve people's lives. However, most image-based emotion recognition techniques are flawed, as humans can intentionally hide their emotions by changing facial expressions. Consequently, brain signals are being used to detect human emotions with improved accuracy, but most proposed systems demonstrate poor performance as EEG signals are difficult to classify using standard machine learning and deep learning techniques. This paper proposes two convolutional neural network (CNN) models (M1: heavily parameterized CNN model and M2: lightly parameterized CNN model) coupled with elegant feature extraction methods for effective recognition. In this study, the most popular EEG benchmark dataset, the DEAP, is utilized with two of its labels, valence, and arousal, for binary classification. We use Fast Fourier Transformation to extract the frequency domain features, convolutional layers for deep features, and complementary features to represent the dataset. The M1 and M2 CNN models achieve nearly perfect accuracy of 99.89% and 99.22%, respectively, which outperform every previous state-of-the-art model. We empirically demonstrate that the M2 model requires only 2 seconds of EEG signal for 99.22% accuracy, and it can achieve over 96% accuracy with only 125 milliseconds of EEG data for valence classification. Moreover, the proposed M2 model achieves 96.8% accuracy on valence using only 10% of the training dataset, demonstrating our proposed system's effectiveness. Documented implementation codes for every experiment are published for reproducibility.

  View details for DOI 10.3390/s22218467

  View details for Web of Science ID 000883990900001

  View details for PubMedID 36366164

  View details for PubMedCentralID PMC9654596

• Accuracy Improvement of Vehicle Recognition by Using Smart Device Sensors SENSORS Pias, T., Eisenberg, D., Fernandez, J. 2022; 22 (12)

  Abstract

  This paper explores the utilization of smart device sensors for the purpose of vehicle recognition. Currently a ubiquitous aspect of people's lives, smart devices can conveniently record details about walking, biking, jogging, and stepping, including physiological data, via often built-in phone activity recognition processes. This paper examines research on intelligent transportation systems to uncover how smart device sensor data may be used for vehicle recognition research, and fit within its growing body of literature. Here, we use the accelerometer and gyroscope, which can be commonly found in a smart phone, to detect the class of a vehicle. We collected data from cars, buses, trains, and bikes using a smartphone, and we designed a 1D CNN model leveraging the residual connection for vehicle recognition. The model achieved more than 98% accuracy in prediction. Moreover, we also provide future research directions based on our study.

  View details for DOI 10.3390/s22124397

  View details for Web of Science ID 000815882100001

  View details for PubMedID 35746179

  View details for PubMedCentralID PMC9228882

• On age prediction from facial images in presence of facial expressions INTERNATIONAL JOURNAL OF APPLIED PATTERN RECOGNITION Shawon, M., Biswas, S., Arefin, N., Pias, T., Rahman, A. 2021; 6 (4): 345-369

  View details for DOI 10.1504/IJAPR.2021.118918

  View details for Web of Science ID 000718028100005

• Human Attention Recognition with Machine Learning From Brain-EEG Signals Hassan, R., Hasan, S., Hasan, M., Jamader, M., Eisenberg, D., Pias, T. edited by Meen, T. H. IEEE. 2020: 16-19

  View details for Web of Science ID 000636949400005

• Gender Recognition by Monitoring Walking Patterns via Smartwatch Sensors Pias, T., Kabir, R., Eisenberg, D., Ahmed, N., Islam, M. edited by Meen, T. H. IEEE. 2019: 220-223

  View details for DOI 10.1109/ecice47484.2019.8942670

  View details for Web of Science ID 000524691500060

• Vehicle Recognition Via Sensor Data From Smart Devices Pias, T., Eisenberg, D., Islam, M. edited by Meen, T. H. IEEE. 2019: 96-99

  View details for DOI 10.1109/ecice47484.2019.8942799

  View details for Web of Science ID 000524691500028

• On the Performance Analysis of APIs Recognizing Emotions from Video Images of Facial Expressions Bhattacharjee, A., Pias, T., Ahmad, M., Rahman, A. edited by Wani, M. A., Kantardzic, M., Sayedmouchaweh, M., Gama, J., Lughofer, E. IEEE. 2018: 223-230

  View details for DOI 10.1109/ICMLA.2018.00040

  View details for Web of Science ID 000463034400032