Feature Selection Using Improved Teaching Learning Based Algorithm on Chronic Kidney Disease Dataset

Feature selection plays an important role in almost any data mining application especially in medical data mining to solve the problem of 'curse of dimensionality' and provide early diagnosis with relevant features and high accuracy. Innumerable feature selection methods have been presented in state-of-arts literature to tackle the problems of high dimensional data. Many evolutionary and swarm intelligence algorithms find solutions based on algorithm-specific control parameters. However, it is a challenging task to identify the optimal feature subset using a feature selection algorithm that is not dependent on the controlling parameters of an algorithm that is specific to a particular problem in hand. Hence, the present research work is based on the working principle of the original TLBO algorithm which does not require any algorithm-specific parameters. The proposed research work is known as Improved Teacher Learner Based Optimization (ITLBO) algorithm which aims to select the best feature subset based on Chebyshev distance formula in the evaluation of the fitness function and common control parameters viz., population size and number of generations to find the optimal feature subset for early diagnosis of chronic diseases. The proposed feature selection technique was applied to Chronic Kidney Disease (CKD) dataset and has achieved a significant feature reduction of 36% compared to the feature reduction of 25 % obtained by applying the original TLBO algorithm. The derived optimal feature subset obtained from TLBO algorithm and feature subset obtained from ITLBO algorithm is validated by evaluating the accuracy of Support Vector Machine (SVM), Convolution Neural Networks (CNN) and Gradient Boosting classification algorithms. Experimental results reveal that there is an overall improvement of classification accuracy for the three algorithms for the derived feature subset from the proposed feature selection algorithm compared to the original TLBO algorithm.