A Repairing Artificial Neural Network Model-Based Stock Price Prediction

Predicting the stock price movements based on quantitative market data modeling is an open problem ever. In stock price prediction, simultaneous achievement of higher accuracy and the fastest prediction becomes a challenging problem due to the hidden information found in raw data. Various prediction models based on machine learning algorithms have been proposed in the literature. In general, these models start with the training phase followed by the testing phase. In the training phase, the past stock market data are used to learn the patterns toward building a model that would then use to predict future stock prices. The performance of such learning algorithms heavily depends on the quality of the data as well as optimal learning parameters. Among the conventional prediction methods, the use of neural network has greatest research interest because of their advantages of self-organizing, distributed processing, and self-learning behaviors. In this work, dynamic nature of the data is mainly focused. In conventional models the retraining has to be carried out for two cases: the data used for training has higher noise and outliers or model trained without preprocessing; the learned data has to update dynamically for recent changes. In this sense, propose a self-repairing dynamic model called repairing artificial neural network (RANN) that correct such errors effectively. The repairing includes adjusting the prediction model from noise, outliers, removing a data sample, and adjusting an attribute value. Hence, the total reconstruction of the prediction model could be avoided while saving training time. The proposed model is validated with five different real-time stock market data and the results are quantified to analyze its performance.