Health Data Classification using Applied Cascade Generalization

Abstract

This research study introduces two steps to improve the binary classification technique without using threshold value. The first step is to use complementary neural networks to produce the truth data and falsity data. The truth and falsity data are used for decision making instead of threshold value. The second step is to separately improve these two data using the applied cascade generalization. The improved truth and falsity data will be used to obtain classification results. The proposed technique is applied to health-related datasets from the UCI machine learning repository which are heart failure clinical records dataset, early stage diabetes risk prediction dataset, and blood transfusion service center dataset. The proposed two-steps technique is found to provide the most accurate average results compared to existing techniques.