Data Mining Techniques in Stock Market Prediction Sen Jiao EECS 435, Data Mining Apr. 14, 2015 Case Western Reserve University
Outline Technical & Fundamental Analysis Bayesian Probability Dynamic Time Series Artificial Neural Network Training
Technical & Fundamental Analysis No black swan events Technical Indicators
Bayesian Probability Update the probability estimates for a hypothesis once additional evidence is learned Stand for performance accuracy of individual stock over a certain period of time Provide standard of optimal decision-making for selecting significant technical indicators
Bayesian Probability 300 trading days Candidate Indicators: MA, Bias, ADX
Dynamic Time Series
Dynamic Time Series
Artificial Neural Network (ANN) Training algorithm iteratively adjusts the connection weights Generalize relevant output when network is adequately trained Training automatically stops when generalization stops improving
Artificial Neural Network (ANN) ANN is expected to yield better prediction results than dynamic time series in most cases # of hidden neurons: 10 70% training data, 15% validation, 15% testing
Preliminary Results Stock: Apple (AAPL) Data from Apr. 11, 2013 to Apr. 11, 2015 505 trading days 450 days training, 55 days prediction Matlab Neural Network Toolbox
Preliminary Results
Preliminary Results – Dynamic DT
Preliminary Results – ANN
Future Work Model Implementation Investigation on more stocks Statistical analysis