Outliers Detection for Regression using K-Means and Expected Maximization Methods in Time Series Data

The evolution of computing technology and the ever increasing size and variety of data sets have created a new range of problems and challenges for data analysts, as well as new opportunities for intelligent systems in data analysis. This study concentrates on performing experimental analysis to find regression base outlier and influential point using two standard algorithms for data clustering are expectation maximization (EM) and K-means. BSESensex time series data has been considered for clustering and outlier detection. The parameters considered in evaluating the results of findings are the number of iterations, the computation time and the memory space consumed at the point of convergence of both K-means and Expectation-Maximization algorithms respectively. Outlier and influential points were detected. The results obtained revealed that Expectation-Maximization algorithm’s quick and premature convergence cannot be said to have guaranteed optimality of results while K-means was found not to guarantee convergence.