Cost Harmonization LightGBM-Based Stock Market Prediction

Stock market prediction (SMP) is a challenging task due to its uncertainty, nonlinearity, and volatility. Machine learning models, such as artificial neural networks (ANNs) support vector regression (SVR), have been widely used for stock achieved high performance in the sense of "minimum errors." In context SMP, however, it more meaningful measure using cost." For example, false positive error (FPE) could result big trading loss, while negative (FNE) might just miss chance. "cautious" investor, fewer FPEs are preferable. fact, cost-sensitive has areas fraud detection medical diagnosis. our earlier study, we proposed false-sensitive method called focal-loss LightBGM (FL-LightGBM) SMP by introducing cost-aware loss LightGBM, which known be fast efficient gradient-boosting algorithm solving large-scale problems. FL-LightBGM, still assumes that all errors (or errors) contribute equally final cost. Such learned strategies useful only an investor who always "aggressive" or "cautious." practice, some may irreversible so important cost based on "data" rather than investor’s character. this paper, propose new cost-harmonization loss-based LightGBM (CHL-LightGBM), each datum can calculated dynamically difficulty datum. To verify effectiveness CHL-LightGBM, comparisons made among XGBoost, decision trees, FL-LightGBM, CHL-LightGBM predictions data from Shanghai, Hong Kong, NASDAQ Exchanges. The simulation results show although there no significant difference between other models accuracy winning rate, obtained highest annual return test data.