Multidimensional Decision Support Indicator (mDSI) for Time Series Stock Trend Prediction

Abstract

This work proposes a generalized approach for predicting trends in time series data with a particular interest in stocks. In this approach, we suggest a multidimensional decision support indicator mDSI derived from a sequential data mining process to monitor trends in stocks. Available indicators in the literature often fail to agree with their predictions to their competitors because of the specific nature of features each one uses in their predictions like moving averages use means, momentums use dispersions, etc. Then again, choosing a best indicator is a challenging and also expensive one. Thus, in this paper, we introduce a compact, but robust indicator to learn the trends effectively for any given time series data. That is, it introduces a simple multdimensional indicator such as mDSI which integrates multiple decision criteria into a single index value that to eliminate conflicts and improve the overall efficiency. Experiments with mDSI on the real data further confirm its efficiency and good performance.