Prediction plays an important role everywhere particularly in business, technology and many others. It helps organizations to take timely decisions, to improve profits and to reduce lost sales. Recent years have witnessed an enormous development in the area of cloud computing and big data, which brings up challenges in decision making process.  As the size of the dataset becomes extremely big, the process of extracting useful information by analysing these data has also become tedious.  Today data are generated in an unprecedented manner, prediction plays major role in utilizing these data. Time Series based prediction models take great part in handling Big Data such as online sales data, weather data etc. In this paper a methodology for prediction is introduced and the model is evaluated by applying various time series models with time series data which is seasonal and non-stationary. From the analysis it is proved that Holt-Winter’s model performs better in seasonal and non-stationary time series data. The Holt-Winters (HW) methods estimate three smoothing parameters, associated with level, trend and seasonal factors. The seasonal variation can be of either an additive or multiplicative form. Also in this paper, Performance Improved Holt-Winters (PIHW) prediction algorithm is proposed and the results demonstrate that a considerable reduction in forecast error (Mean Square Error) can be achieved in the proposed model compared to Holt-Winters (HW) model.

Keywords: Big Data,  Holt-winter, Hadoop,  MapReduce, Prediction