Forex prediction engine: framework, modelling techniques and implementations

Having accurate prediction in foreign exchange Forex market is useful because it provides intelligent information for investment strategy. This paper studies extracted repeating patterns of historical Forex time series, so to predict future trend direction by matching the forming trend with a repeating pattern. In the proposed Forex prediction engine, global pattern movements over a period of time are extracted using a linear regression line LRL enhanced technique, and then further segmented into what we called up and down curves. Subsequently, the artificial neural network ANN is applied to classify or group the uptrend and downtrend patterns. Finally, the dynamic time warping DTW is used through brute force to identify a trend pattern similar to the current trend at least for the beginning part. The remaining part of the matched pattern can provide predictive clues about next day trend movement. The experimental results generated on the dataset of AUD-USD and EUR-USD currencies between 2012 and 2013 demonstrate reliable accuracy performance of 72%.