Home / Current Issue / Paper 1708714
Machine Learning-Based Forecasting of Financial Time Series Using Ensemble Techniques and Technical Indicators
Subject area: Science,Engineering and Technology · Area of research: AI and ML in Finance
Abstract
Financial markets present a highly dynamic and nonlinear environment, making accurate time series forecasting both challenging and critical. On the other hand, investors are highly concerned about the future stock price trend. There are still challenges in the stock price prediction model, including the extensive number of methods, the selection of feature index variables, and the substantial time-effectiveness of prediction. This study aims to propose a machine learning-driven approach for stock return prediction and trading signal generation using an ensemble of XGBoost and Logistic Regression models. First, we denoise the raw data with statistical-based strategies. Second, we engineered a comprehensive feature set incorporating lagged returns, technical indicators including RSI, SMA, MACD, and volatility. Strategies were evaluated using walk-forward validation to ensure realistic simulation of evolving market behavior. Next, trade signals were filtered with RSI thresholds and capped with a stop-loss mechanism. Finally, performance was measured using risk-adjusted metrics including Sharpe, Sortino, Calmar ratios, and maximum drawdown. The results show that the ensemble strategy outperformed single-based models and traditional moving average crossover methods in both cumulative return and risk stability.
Keywords
Dynamic pricing, Data-driven architecture, Machine learning, XGBoost, logistic regression
How to cite this paper
@article{1708714,
author = {Merry Jebin Gnanadhas},
title = {Machine Learning-Based Forecasting of Financial Time Series Using Ensemble Techniques and Technical Indicators},
journal = {Iconic Research And Engineering Journals},
year = {2025},
volume = {8},
number = {11},
pages = {1791-1802},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1708714.pdf},
abstract = {Financial markets present a highly dynamic and nonlinear environment, making accurate time series forecasting both challenging and critical. On the other hand, investors are highly concerned about the future stock price trend. There are still challenges in the stock price prediction model, including the extensive number of methods, the selection of feature index variables, and the substantial time-effectiveness of prediction. This study aims to propose a machine learning-driven approach for stock return prediction and trading signal generation using an ensemble of XGBoost and Logistic Regression models. First, we denoise the raw data with statistical-based strategies. Second, we engineered a comprehensive feature set incorporating lagged returns, technical indicators including RSI, SMA, MACD, and volatility. Strategies were evaluated using walk-forward validation to ensure realistic simulation of evolving market behavior. Next, trade signals were filtered with RSI thresholds and capped with a stop-loss mechanism. Finally, performance was measured using risk-adjusted metrics including Sharpe, Sortino, Calmar ratios, and maximum drawdown. The results show that the ensemble strategy outperformed single-based models and traditional moving average crossover methods in both cumulative return and risk stability.},
keywords = {Dynamic pricing, Data-driven architecture, Machine learning, XGBoost, logistic regression},
month = {May},
}