Machine Learning Model for Attenuating Outliers in Stock Data

The presence of outliers has deleterious effects on the stock value because the unreliable information may discourage investors from investing in the stock. This associated problem paved way for the importance of intelligent prediction paradigm. However, existing stock forecasting models like Artificial neural network (ANN) performed better than traditional statistical models in handling the problem of non-linearity and complexity in forecasting stock price, it still lacked the capacity to handle outliers which are inherent in the stock market. Based on this, the researchers were motivated to forecast stock market based on the observation from literature that researchers do not report checking, prediction and proper management of outliers of any sort in stock market price forecasting. This paper addressed outliers' deleterious effects on the stock value by proposing a hybrid Fuzzy Artificial Neural Network (FANN) Model that attenuates outliers in stock forecasting accurately. The proposed model was simulated using MATLAB. The historical Nigerian stock quantitative datasets from Nigeria Stock Exchange (NSE) for 2008-2011 were used to test the simulated model. The Gaussian Membership function, due to its ability to handle minimum uncertainty principle, was used for the fuzzification of the extracted stock features to capture the stock dynamism. The proposed model's predictive performance was calculated using Root Mean Square Error (RMSE). The outlier detection analysis of the actual historical stock data, ANN and the proposed FANN Model predictions was calculated using Z-score. The proposed model had a RMSE value of about 3.83 which shows that it is a reliable stock forecasting model. The Z-score value of the proposed model was calculated to be about 0.78 which shows that it significantly attenuates outliers in stock forecasting. In overall, the results proved that the proposed FANN model can handle outliers in stock forecasting.