Diabetes Prediction Tool under System on Chip Using Machine Learning Method

Abstract

Extraordinary advances in biotechnology and health sciences have brought significant generation of data, such as genetic data and clinical information, generated from huge electronic health records. Data analysis is a process of studying and identifying hidden patterns from large amounts of data and drawing conclusions. In health care, this analytical process is carried out using machine learning (ML) algorithms to analyze medical data to build Machine Learning models to transform all available information into valuable knowledge. Nowadays, diabetes has become a common disease among young people, elderly people, and even children. According to the World Health Organization (WHO) report, by 2025, this number is expected to exceed 380 million. This research work performs a comparison of 5 classification techniques, namely Naive Bayes (NB), Bagging (J48), Decision Tree (J48, C4.5), K Nearest Neighbors (KNN), and Support Vector Machine (SVM), to detect diabetes at an early stage. The performances of the five algorithms are evaluated and compared on various measures like accuracy, precision, and recall. The experiments were conducted based on the diabetes database, the source from the National Institute of Diabetes, Digestive and Kidney Diseases, showing the effectiveness of using the Decision Tree (DT) technique. This effectiveness led to the choice of this method. After obtaining the DT model of the problem, the main task facing us in this work is the hardware implementation of this model. Indeed, this forecasting system can be used in other complementary works as a processing unit in a cloud, to be able to manage numerous requests. The considered solution is the hardware implementation on a Field Programmable Gate Array (FPGA) board.