Forecasting electronic industry EPS using an integrated ANFIS model

The process of buying stock, the major indicator is earning per share (EPS). It is the earning return on original investment; it represents the profit ability of common stock, and the final result of company performance. Therefore, in this study integrates financial-statement related indicators to predict the future EPS. Base on literatures, the relationship of EPS and related attributes is nonlinear, and the nonlinear model can predict well in EPS, so we propose an integrated adaptive network-based fuzzy inference system (ANFIS). It combines with the decision tree which is the pre-process for enhancing predicting ability, and there are three stages in study (1) use feature selection to reduce attributes, and the attributes are discretized by decision tree, then encoding the dicretization value (2) take fuzzy inference system (FIS) to fuzzify the encoding value, and use adaptive network to tune optimal parameters. (3) employ an integrated ANFIS model to predict EPS. We collect nine-quarter EPS data for predicting, and then the proposed method surpasses in accuracy these conventional data mining models.