H³O-LGBM: hybrid Harris hawk optimization based light gradient boosting machine model for real-time trading

Area of Science:

• Economics
• Data Science
• Machine Learning

Background:

• India's Gross Domestic Product (GDP) is significantly influenced by its trade and external sector.
• Exports and imports saw substantial year-over-year increases in FY2021-2022.
• Machine learning holds potential to boost India's gross value by 15% by 2035.

Purpose of the Study:

• To present an optimized Light Gradient Boosting Machine (Light GBM) model for trade forecasting.
• To enhance artificial intelligence growth through the integration of data, technology, and talent.
• To predict stock price movements (hold, buy, or sell) by analyzing trade data.

Main Methods:

• Developed a hybrid Harris Hawk Optimization (H3O) algorithm to optimize Light GBM parameters.
• Addressed overfitting in Harris Hawk Optimization using Exclusive Feature Bundling (EFB) and Gradient-based One-side Sampling (GOSS).
• Corrected residual errors of the optimized Light GBM model using a Markov Chain model.

Main Results:

• The H3O algorithm demonstrated fast convergence by optimizing Light GBM parameters.
• The proposed hybrid model effectively extracted crucial input values for trade variables.
• Evaluated on precious metal and stock market datasets, the model showed enhanced prediction performance and accuracy.

Conclusions:

• The optimized Light GBM model accurately predicts economic growth and market conditions.
• The hybrid approach improves prediction accuracy for Intermarket Trading Systems (ITS) and services.
• The model aids in making informed stock trading decisions based on market analysis.