Machine Learning Model for Diabetes Prediction Using Parallel Computing: Comparative Study

Abstract

Diabetes has become one of the most common and challenging health conditions in the world because it alters how the body uses glucose, an essential source of energy, and can damage organs including the kidneys, heart, eyes, and other issues in addition to the blood. Therefore, it is necessary to develop a system that can accurately identify diabetes patients using medical. indicators. Artificial intelligence (AI)-based techniques such as Machine Learning (ML) have proven to be effective in this regard. Sequential machine learning methods utilize a single underlying hardware processing element, thus having poor real-time prediction efficiency. Moreover, these approaches may struggle to handle large amounts of data due to their time-consuming nature. Parallel computing has been widely applied techniques that utilize multiple hardware processing elements to increase the application's computation time. In this study, we utilized parallel techniques in Python to train ML models and provided a comparative study for different parallel techniques. We used the Pima Indian Diabetes Dataset (PIDD), conducted five different experiments, and provided a comparative performance evaluation. We deployed two ML models which are Decision Tree (DT) and Linear Regression (LR). For each model, we compared the sequential execution with three different parallel Python techniques (multithreading, multiprocessing, and loky), each utilizing four cores. Our results showed that LR with multiprocessing technique achieved a higher accuracy of 78% and greater speedup of 39. The results in general indicated that parallel execution outperforms sequential execution in terms of speed. This comparative study provides valuable insights into how to optimize machine learning models for diabetes detection and highlights the usefulness of parallel computing technologies in healthcare applications.