Hybrid Feature Selection Method for Health Data Mining

Abstract

Background: Feature selection is one of the most important parts of machine learning for predicting the outcome. There are methods for selecting features or generating feature subset. Such as: Filter method, wrapper method. Previously features were selected using any one of these two methods. The result from the method was pretty good but it could be better. Objective: The objective of my thesis is to get the more accurate result. Here I am emphasizing on feature selection for getting the more accurate results. There is another method for feature selection, which is: hybrid method. Hybrid method combines both the filter and wrapper methods. Here I am going to use the hybrid method for selecting features and I will show that hybrid method can get the same or more accurate result using less features. Results: The final result is showing that, in some cases hybrid method is giving the same results as filter and wrapper method. Some other cases show that hybrid method is giving more accurate result that filter and wrapper method. In all the cases hybrid method is using less features than filter and wrapper method. Here filter and wrapper method is using four features each and hybrid method is three features.