Mahmudul Islam, Syed Tangim Pasha, Jahanggir Hossain Setu, Nabarun Halder, Ashraful Islam, M. Ashraful Amin
2024 7th Asia Conference on Cognitive Engineering and Intelligent Interaction (CEII 2024)
IEEE, pp. 1-5
This study examines the increasing incidence of diabetes in Bangladesh and the necessity for effective tools to categorize and forecast diabetic situations utilizing Machine Learning (ML) methodologies. It compares the performance of predictive models on the DiaHealth dataset and the widely used PIMA Indians Diabetes dataset. Two ML algorithms, Random Forest (RF) and Naïve Bayes (NB), have been used to classify diabetic and non-diabetic individuals. This study used the PIMA Indians Diabetes dataset (768 samples) and the Dia-Health dataset from Bangladesh (5,437 samples). Both datasets contain attributes including insulin levels, blood pressure, Body Mass Index (BMI), and age. Models have been evaluated using performance metrics such as accuracy, precision, recall, F1-score, and Receiver Operating Characteristics-Area Under the Curve (ROC-AUC). The RF model achieved 93.93% accuracy on the DiaHealth dataset and 73.38% on the PIMA Indians Diabetes dataset. In contrast, NB performed better on non-diabetic patients in both datasets but struggled to accurately classify diabetic patients, particularly in the DiaHealth dataset. The models showed variance in performance across datasets. This indicates demographic differences and feature distributions. The study shows that RF is effective in predicting diabetes in local Bangladeshi populations but also faces challenges with class imbalance and feature variation. This study also indicates the need for dataset-specific tuning of ML models in diabetes classification and prediction tasks.
Assistant Professor
Department of Computer Science and Engineering
Independent University, Bangladesh (IUB)Human-Computer Interaction, AI for Social Good, AI for Public Health, AI for Impact