Early Stage Diabetes Risk Prediction via Machine Learning

Abstract

AbstractDiabetes disease is triggering many health problems including microvascular diseases, macrovascular abnormalities, and neuropathic diseases. From an economic perspective, diabetes is one of the costliest diseases, moreover, high percentage of adults with diabetes live in low- and middle-income countries triggering more economic troubles to these countries. Diagnosing the risk of diabetes will help combat against this silent killer. In this paper, we propose an inclusive machine learning based predictive model for diagnosing the risk of having diabetes using a recent dataset of signs and symptoms, known as Diabetes Risk Prediction (DRP2020). We employ more than twenty ML techniques on DRP2020 and we evaluate all ML based models using different performance evaluation metrics including accuracy, precision, recall, harmonic mean, prediction speed and alarm errors. We provide extensive simulation results and compare the performance of various ML based models. Accordingly, the model based shallow neural networks (SNN) has been elected as the optimum model for constructing of the early stage diabetes risk prediction scoring a 99.23% and 99.38% for prediction accuracy and the harmonic mean of precision and recall, respectively. The obtained results exhibit the proficiency and distinction of our model over other state-of art models. Eventually, the proposed system can be proficiently deployed as a clinical tool to assist in provide early stage diabetes risk prediction.KeywordsMachine learningDetectionDiabetes diseaseRisk predictionAccuracy