A DEEP LEARNING INSPIRED LIVER DISEASE PREDICTION MODEL WITH OPTIMAL WEIGHTED RBM KERNELS FOR FEATURE EXTRACTION AND HYBRID DEEP RESIDUAL NETWORK

Abstract

Since liver disease often presents no clear symptoms in its early stages, misdiagnosis by doctors can occur, putting patients' health at risk. Generally, liver disease does not show early-stage symptoms to the false diagnosis, which may lead to wrong treatment and cause a threat to the health of the patient. However, the traditional methods are very complex and expensive. Deep learning techniques are employed to solve challenges, enabling effective treatment of liver disease by leveraging vast amounts of data. Initially, the patient data on various liver diseases are collected for the dataset. The data is sent to data cleaning process for removing unwanted data. Then, the cleaned data is sent for accurate feature extraction, and the Weighted Restricted Boltzmann Machine (WRBM) kernels are considered for identification of the high dimensional features. Also, here the weights involved in the process are optimally generated by the newly developed Intellectual and Randomized Integer Amendment-based Starfish Optimization Algorithm (IRIA-SOA). Then, high dimensional features are trained using the proposed Hybrid Deep Residual Network (HyDRNet), which is a combination of both Convolution Extreme Learning Machine (Conv-ELM) and Echo State Network (ESN). The presented network can process large patient data records, and predict different liver disease types.