Forecasting the Risk of Type II Diabetes using Reinforcement Learning

Type II Diabetes (T2D) is one of the most common lifestyle diseases which is characterized by insulin resistance. Lack of insulin's proper working causes uncontrollable blood glucose rise in the body which leads to life taking situations. Therefore, early detection of T2D is imperative to save many lives. Towards this goal, this work presents a machine learning-based prediction model to detect T2D. The Q-learning algorithm belonging to the Reinforcement Learning (RL) paradigm has been applied to the PIMA Indian Women diabetes dataset in developing the detection model. The model identifies patients with T2D using three factors (such as Body Mass Index, glucose level and age of subject) by generating an off-policy based RL and making the learning agent to find an optimal policy for the factors. The information of a subject can be in any of 330 possible states. The proposed RL model's accuracy, Precision, Recall, F-measure and AUC values have been compared with the state-of-the-art techniques such as K Nearest Neighbors and Decision Tree. The performance of the proposed RL-based T2D prediction outperforms the K Nearest Neighbors and Decision Tree.