Neural Network-Based Intelligent System for Glucose Spike Prediction and Type 2 Diabetes Mellitus Risk Assessment

Machine learning algorithms have revolutionized the management of clinical data to prevent Type 2 Diabetes Mellitus (T2DM). This paper performs an exploratory study of the relationship between several risk factors, including visceral fat percentage (VF), sedentary lifestyle (SL), sleep hours, adherence to the diet, genetic predisposition (GP), sex, body mass index (BMI), fasting plasma glucose levels (FPG), waist circumference (WC), low dietary fiber intake (LDFI), high carbohydrate intake (HCI), and sleep hygiene, with the development of T2DM. For it, a chi-square statistical tool was used to discern the significance of these features concerning T2DM incidence; among these, VF, SL, sleep hours, and adherence to the diet demonstrated the highest significance. To predict the development of T2DM, a neural network (NN) classifier is proposed, which achieved remarkable accuracy (100%) during the training phase. A Levenberg-Marquardt supervised learning algorithm was employed to minimize the total error. Furthermore, Long-Short Term Memory (LSTM) is leveraged to forecast glucose spike generation, enabling a comprehensive estimation over a seven-day period. The accuracy of the prediction was validated through the Root Mean Square Error (RMSE), which resulted in a value of 0.5803. Clinical Relevance—This exploratory study emphasizes the importance of risk factors and preventive medicine in the context of T2DM. By shifting focus towards prevention, clinicians can transform patient care and promote better health outcomes. Leveraging the combined power of risk factor analysis and machine learning techniques empowers clinicians to effectively guide individuals at risk of developing T2DM. This approach facilitates informed lifestyle choices, potentially leading to a significant reduction in the overall disease burden. This research holds significant promise for improving population health and transforming healthcare practices towards a preventive model