Machine learning techniques for short term stock movements classification for Moroccan stock exchange

Accurate stock price forecasting is important for investors and traders to make informed trading decision. However, prices have a complex behavior due to their nonlinearity and nonstationarity. In this paper three Machine learning techniques are implemented to predict a very short term (10 minutes ahead) variations of the Moroccan stock market: Random Forest (RF), Gradient Boosted Trees (GBT) and Support Vector Machine (SVM). A selection of technical indicators was used as inputs variables and a feature selection and samples selection steps were performed to improve prediction accuracy and training time. An eight-year period of intraday prices (tick-by-tick data) of Maroc Telecom (IAM) stocks is employed as experimental database to evaluate the performances of the selected models. The experimental results have shown that RF and GBT are superior to SVM for our dataset. Further, the low computational complexity and reduced training time of RF and GBT are suitable for short term forecasting.