Unsupervised Learning on Healthcare Survey Data with Particle Swarm Optimization

Abstract

The behavior of a machine learning algorithm very much depends upon the structure of entities. There a pool of big data, which have lots of ambiguity, noises, granularity, unlabeled behavior that make it tough to find pattern and visualize an outcome. To solve this issue machine learning proposes an algorithm like clustering, which is mainly part of unsupervised learning. In the proposed model we used the unsupervised learning specifically the clustering algorithms like Expectation Maximization (EM), Make Density Based Analysis (MDBC) wrapped with the clustering algorithm like Kmeans, EM, GenClust++. The dataset has been chosen is a survey data on healthcare issues like chronic diseases, regular treatment etc. across various states and district of India. We have used Weka tool for formulation and training our dataset. This dataset is obtained from the Open Data Government Platform India Portal under the Government of Open Data License India, from the Department of Health and Family Welfare. The dataset contains ordinal type of values. These kinds of dataset are excellent examples of exploratory data analysis. The particle swarm optimization is applied on dataset for optimizing the attributes later trained with clustering algorithms which gave better log likelihood value comparable to the results on simple clustering algorithms. We even compare the performance speed among the algorithms to check how quickly they are able to perform on the variety of attributes.