Multi disease prediction based on combined deep reinforcement Boltzmann machines
Vetrithangam, Aruna Devi, Shruti Aggarwal
Abstract
Machine learning (ML) and deep learning (DL) play a decisive role in increasing the performance of activities like risk prediction of diseases in the health care system, allowing societies to be better served. The authors present a method for multi disease prediction combining the deep reinforcement method and improved the Boltzmann machine method. The proposed system uses the deep learning algorithm that is the deep reinforcement algorithm, and the improved the Boltzmann machine method to ascertain the strongest and most effective features which give the best in terms of disease prediction accuracy. Deep Reinforcement Boltzmann machines extract features from raw data or datasets in layers, with the features retrieved in one layer being used as hidden variables as input to the next layer. The process of pre-training is performed on the unlabeled data in a layer wise manner and the training algorithm is fine-tuned. The optimal behavior/features of the environment’s model are acquired by doing actions and noticing the results, that contains the next condition, as well as the immediate gratification. The disease prediction is accomplished based on the features extracted from the raw dataset. Many diseases such as retinal fundus multi-disease, diabetic retinopathy, skin cancer disease detections and stages predictions are considered in disease prediction. We have compared performances of deep learning techniques when applied to different data sets beyond numerous application domains and the accuracy is improved from 98.0 % to 98.60% and the precision is improved from 98.2% to 99.2%.