Predicting Trend of High Frequency CSI 300 Index Using Adaptive Input Selection and Machine Learning Techniques

High-frequency stock trend prediction using machine learners has raised substantial interest in literature. Nevertheless, there is no gold standard to select the inputs for the learners. This paper investigates the approach of adaptive input selection (AIS) for the trend prediction of high-frequency stock index price and compares it with the commonly used deterministic input setting (DIS) approach. The DIS approach is implemented through computation of technical indicator values on deterministic period parameters. The AIS approach selects the most suitable indicators and their parameters for the time-varying dataset using feature selection methods. Two state-of-the-art machine learners, support vector machine (SVM) and artificial neural network (ANN), are adopted as learning models. Accuracy and F-measure of SVM and ANN models with both the approaches are computed based on the high-frequency data of CSI 300 index. The results suggest that the AIS approach using t-statistics, information gain and ROC methods can achieve better prediction performance than the DIS approach. Also, the investment performance evaluation shows that the AIS approach with the same three feature selection methods provides significantly higher returns than the DIS approach.