Scholarlink

Financial markets are characterized by volatility and complex non-linear dependencies, posing a significant challenge for reliable stock recommendation. This paper introduces SafeTrade-A, a novel hybrid machine learning framework that integrates Gradient-Boosted Decision Trees (GBDT) with quantile regression to deliver risk-aware recommendations. Utilizing historical S&P 500 data (2010–2024, 3.5 M records), our method synthesizes expected return prediction with downside risk estimation to compute a composite risk-adjusted score for ranking stocks. Experimental evaluation demonstrates that SafeTrade-A achieves superior predictive performance (RMSE: 0.0205, R2: 0.68) and delivers higher cumulative returns (38.2%) with substantially reduced maximum drawdown (12.5%) compared to GBDT-only, LSTM, and ARIMA benchmarks. Statistical significance testing confirms the robustness of these improvements (p < 0.01). Furthermore, SHAP-based interpretability analysis enhances transparency, identifying momentum, volatility, and sector ETFs as key predictive drivers. The results affirm that SafeTrade-A constitutes an effective, explainable, and risk-sensitive recommender system for informed trading decisions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Safe Trade-A Stock Recommender Using Machine Learning Algorithms

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • H. B. Ranjitha,
  • E. Santosh

摘要

Financial markets are characterized by volatility and complex non-linear dependencies, posing a significant challenge for reliable stock recommendation. This paper introduces SafeTrade-A, a novel hybrid machine learning framework that integrates Gradient-Boosted Decision Trees (GBDT) with quantile regression to deliver risk-aware recommendations. Utilizing historical S&P 500 data (2010–2024, 3.5 M records), our method synthesizes expected return prediction with downside risk estimation to compute a composite risk-adjusted score for ranking stocks. Experimental evaluation demonstrates that SafeTrade-A achieves superior predictive performance (RMSE: 0.0205, R2: 0.68) and delivers higher cumulative returns (38.2%) with substantially reduced maximum drawdown (12.5%) compared to GBDT-only, LSTM, and ARIMA benchmarks. Statistical significance testing confirms the robustness of these improvements (p < 0.01). Furthermore, SHAP-based interpretability analysis enhances transparency, identifying momentum, volatility, and sector ETFs as key predictive drivers. The results affirm that SafeTrade-A constitutes an effective, explainable, and risk-sensitive recommender system for informed trading decisions.