TITLE:
Multi-Forest Trading Algorithm: A Novel Framework for Equity Price Prediction Using Disrupted Time-Series Data
AUTHORS:
Jaideep Padhi, Clayton Greenberg
KEYWORDS:
Stock Price Prediction, Machine Learning, Time-Series Forecasting, Binary Classification, Disrupted Time-Series Data
JOURNAL NAME:
American Journal of Industrial and Business Management,
Vol.15 No.5,
May
31,
2025
ABSTRACT: The application of artificial intelligence in stock price forecasting is an important area of research at the intersection of finance and computer science, with machine learning techniques aimed at predicting future price movements, seeking consistent and profitable financial outcomes. However, due to the volatile nature of financial markets, generating predictions regarding an equity’s future performance is challenging due to the complex and diverse factors that influence stock price dynamics, particularly concerning intraday movements. Prior research primarily focuses on hyperparameter optimization, feature engineering, and hybridization, often overlooking fundamental modifications to model and data architecture. Altering model and data architecture during model creation can significantly enhance model performance under real-time market conditions, to a greater extent than the aforementioned methods. Time-series forecasting in equity prices consists of two dimensions: magnitude and direction. Current algorithms used for stock price prediction have reduced efficiency as they attempt to forecast both dimensions with a single model. This paper introduces the Multi-forest model, a novel approach to stock price prediction that implements a bilayer machine learning algorithm combining sequential binary classification processes and regression processes to increase prediction accuracy. Although the classification process disrupts the continuity of the time-series data, the regressor effectively generates valid predictions, dispelling notions that a complete time-series is required for accurate predictions. The Multi-forest Trading Algorithm (MTA) demonstrates effectiveness during temporal deployments, providing success rates of 93.4%, 94.1%, and 84.0% for April 2024, June 2024, and October 2024, respectively, months differing greatly in volatility and overall performance. When compared to models currently implemented in stock price prediction, the MTA outperformed all by a minimum margin of 15%, providing consistent results and exhibiting cautionary behavior when faced with volatile market conditions. Regarding profitability, the algorithm produced profit factors of 28.8, 31.0, and 15.0 for each respective month in the temporal deployments, indicating a projected profitability between 5 - 7 times greater than that of current algorithms.