An Ensemble of Feature Subset Selection with Deep Learning Based Data Classification Model for Healthcare Diagnosis System
Due to advanced developments in database technologies, it is simpler to save the medical records of hospital patients from their initial day of admission compared to earlier days. Intensive Care Units (ICU) in the advanced medical information system records the patient details in relational databases at every instant of time. Given a collection of electronic patient records, a predictive model which efficiently allocates the disease labels can help medical database management and assist physicians. At the same time, Feature selection is commonly employed to reduce the number of features to increase predictive results, better visualization, comprehensibility and reduce training time. In this view, this paper presents a new deep learning based predictive model for disease diagnosis to assist physicians using well-known disease code system called International Statistical Classification of Diseases and Related Health Problems (ICD). A new simulated annealing (SA) based feature selection (FS) with long short term memory (LSTM) based disease classification model called SA-LSTM has been presented. The SA-LSTM model initially undergoes pre-processing, followed by FS and classification. An extensive set of experimental analysis takes place on three dataset and the obtained resulted indicated the effective performance of the SA-LSTM model with the maximum accuracy values of 95.48%, 84.49% and 79.36% on the applied Diabetes, EEG Eye State and Framingham dataset respectively.