Scholarlink

<p>India’s real-time forecasting of stock prices is hindered by heterogeneous data sources, volatility in markets, and regulatory demands for transparency in algorithmic trading. This research proposes a systematic, interpretable multi-source artificial intelligence (AI) framework for tracking and predicting stock prices on the NSE and BSE. The model brings together market technical indicators, macroeconomic variables, and sentiment factors into one prediction pipeline through a hybrid ensemble of temporal fusion transformers (TFTs), long short-term memory (LSTM) networks, random forests (RFs), and extreme gradient boosting (XGBoost). Deep learning models capture temporal relationships, while tree-based methods provide structural stability. Explainability is ensured through SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and counterfactual analysis, allowing for transparent decision-making as per SEBI guidelines. Results on multi-year NSE/BSE data (2015–2024) show a 39.8% RMSE reduction and a 9.62% increase in directional accuracy relative to baselines employing conventional and single-model methods. Ablation studies discover that macroeconomic and sentiment data significantly enhance short- and medium-term forecasting accuracy, particularly during event-driven volatility. The results highlight the promise of the framework in enabling low-latency, explainable, and regulation-compliant prediction, enabling algorithmic trading, compliance with policy, and enhanced decision-making.</p>

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

Explainable Multi-source AI Framework for Real-Time Stock Price Monitoring and Prediction

  • 期刊论文

  • 发表日期：

    2026年1月21日

查看全文
  • Manikrao Patil,
  • Jaikumar M. Patil,
  • Aniket K. Shahade,
  • Priyanka V. Deshmukh

摘要

India’s real-time forecasting of stock prices is hindered by heterogeneous data sources, volatility in markets, and regulatory demands for transparency in algorithmic trading. This research proposes a systematic, interpretable multi-source artificial intelligence (AI) framework for tracking and predicting stock prices on the NSE and BSE. The model brings together market technical indicators, macroeconomic variables, and sentiment factors into one prediction pipeline through a hybrid ensemble of temporal fusion transformers (TFTs), long short-term memory (LSTM) networks, random forests (RFs), and extreme gradient boosting (XGBoost). Deep learning models capture temporal relationships, while tree-based methods provide structural stability. Explainability is ensured through SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and counterfactual analysis, allowing for transparent decision-making as per SEBI guidelines. Results on multi-year NSE/BSE data (2015–2024) show a 39.8% RMSE reduction and a 9.62% increase in directional accuracy relative to baselines employing conventional and single-model methods. Ablation studies discover that macroeconomic and sentiment data significantly enhance short- and medium-term forecasting accuracy, particularly during event-driven volatility. The results highlight the promise of the framework in enabling low-latency, explainable, and regulation-compliant prediction, enabling algorithmic trading, compliance with policy, and enhanced decision-making.