Supervised learning algorithm with bivariate AR(p) model

Abstract

Time series classification is one of the most predominant factors for efficient data analysis in finance, image processing, signal processing, medical imaging, speech recognition etc. Much work has been reported in literature regarding supervised learning algorithm in time series data. In all these algorithms, it is customary to consider each variate separately for mining bivariate time series (Xiong and Yeung, Pattern Recogn 37:1675–1689, 2004). However, in many practical situations, one variable is highly influenced by other variable and hence analyzing each variable isolately may not provide accurate results. Very little work has been reported in supervised learning algorithm with bivariate auto regressive model of order P. Hence, in this paper, we introduce a new and novel supervised learning algorithm with bivariate auto regressive model of order P to mine the bivariate time series data. It is assumed that the whole data set is characterized by the M-component mixture of Bivariate AR (p) model. The number of components is known and fixed by defining the query under consideration. The model parameters are estimated using EM algorithm. The supervised learning algorithm is developed by utilizing component maximum likelihood and Bayesian information criteria. The performance of the developed algorithm is studied through data collected from opening price stock of a portfolio and closing price stock of a portfolio. A comparative study of the developed algorithm is made with the data mining algorithm with univariate AR(2) model. The Comparative study of developed algorithm reveal that its performance is better than existing algorithms. This algorithm also includes Univariate supervised learning algorithm as a particular case.