Developing a novel stock index trend predictor model by integrating multiple criteria decision-making with an...

Area of Science:

• Computational Finance
• Machine Learning Applications
• Financial Time-Series Forecasting

Background:

• Accurate stock market trend prediction is a long-standing research goal.
• Advanced forecasting models are crucial for predicting stock prices, market fluctuations, and trading profits.
• The performance of Online Sequential Extreme Learning Machine (OSELM) heavily relies on its activation functions.

Purpose of the Study:

• To design a novel stock index trend predictor by integrating Multiple Criteria Decision-Making (MCDM) with an optimized Online Sequential Extreme Learning Machine (OSELM).
• To forecast future stock index prices and analyze their upward or downward trends.
• To address the selection of optimal activation functions for OSELM as an MCDM problem.

Main Methods:

• A novel stock index trend predictor model combining MCDM with an optimized OSELM.
• Selection of the best activation function for OSELM using three MCDM approaches, evaluated against ten criteria (five regression-based, five classification-based).
• Optimization of OSELM through a hybrid crow search algorithm (hCSA) incorporating chaotic maps, mutation operators, and catfish behavior for improved convergence.

Main Results:

• The proposed hCSA-OSELM model demonstrated superior performance over state-of-the-art baseline models on historical data from BSE SENSEX, S&P 500, and DJIA during pre-COVID and COVID periods.
• Significant improvements in Mean Squared Error (MSE) of 4-6% (pre-COVID) and 25-31% (COVID), and accuracy improvements of 0.4-0.8% (pre-COVID) and 0.9-1.3% (COVID) were observed compared to the second-best model.
• Statistical tests confirmed the enhanced performance of the proposed model.

Conclusions:

• The MCDM-based model selection provides a robust and reliable approach for optimizing OSELM.
• The hCSA-OSELM model achieves superior prediction and classification outcomes for financial time-series forecasting.
• The proposed model is effective for navigating both daily volatility and highly volatile market conditions.