The most important classification difficulty is feature selection. Feature selection is widely used data analytics and machine learning tasks and it is one of the data preprocessing tasks. It does not produce new features. It attempts diminish the attributes by removing irrelevant attributes using an efficient and robust feature selection method. If the dataset contains irrelevant attributes it may lead to poor performance of the classifier. A significant method is used to select the appropriate or relevant set of features. In this paper, several feature selection algorithms using evolutionary techniques are presented. Mostly evolutionary techniques are used for finding the relevant features corresponding to the disease.
1 May 2017 in Indian Journal of Science and Technology
The objective of this research is to develop a decision support system for the investigation of type II diabetes and its risk factors over the region of specific group of people. A total of 732 cases were collected from a government hospital. Predictive analysis was carried out using bee based harmony search algorithm and C4.5 decision tree algorithm with its splitting criterion. From the experimental results, it has been observed that the risk corresponding to Postprandial Plasma Glucose (PPG), A1c-Glycosylated Haemoglobin, Mean Blood Glucose level (MBG), with a prediction accuracy of about 92.87% respectively. It is estimated that the age group corresponding to 34 to 73 was found more prevalent to the disease. The mathematical model proves that age, PPG and MBG have strong co-relation over data matrix. Hence predictive analytics with swarm intelligence techniques can be deployed over risk identification which reduces treatment analysis.