TradeSense: An Explainable AI-Based Trading Decision Support System for Retail Investors

Abstract

Individual investors today face a significant imbalance. They have access to vast amounts of data, but the tools available to help them understand it are basic. Most retail trading platforms provide little more than historical charts and a few standard indicators. Users must manually piece together price signals, news events, and portfolio exposure without any structured support. This paper presents TradeSense, a decision support system that aims to bridge this gap by combining three areas of applied AI research: short-term price trend prediction, news-driven sentiment analysis, and retrieval-augmented natural language explanation. This all comes together in a single web application designed for non-institutional investors. TradeSense primarily relies on an XGBoost classifier, which estimates the likelihood of a stock’s price trend continuing over five days. A FinBERT-based component processes recent financial news headlines, generating a sentiment index that users can compare with the quantitative prediction. A Retrieval-Augmented Generation (RAG) engine then combines these outputs into an easy-to-understand explanation, using a curated knowledge base of past analyses and indicator definitions. The system features a three-tier architecture, with a React frontend for user interaction, a Node.js/Express gateway for managing operations, and a Python/FastAPI backend for the analytical tasks. A lightweight portfolio module tracks user holdings and highlights concentration risk. Evaluations through historical backtesting, event-driven case studies, and structured software testing show that this integrated approach makes model outputs much more interpretable than presenting raw signals alone. However, real-world use would require significant improvements in data coverage and infrastructure.

Introduction

TradeSense is an AI-powered stock market analytics platform designed to help retail investors make informed decisions by integrating machine learning, financial sentiment analysis, portfolio analytics, and explainable AI into a single web application. Unlike conventional trading platforms that mainly display historical charts and technical indicators, TradeSense provides short-term stock trend predictions, news sentiment analysis, natural language explanations, and portfolio risk insights. The system employs an XGBoost classifier to predict five-day stock price movements using technical indicators, FinBERT to analyze the sentiment of financial news, and a Retrieval-Augmented Generation (RAG) framework to generate transparent, evidence-based explanations for predictions. A portfolio management module further calculates unrealized profit/loss, asset allocation, and concentration risk, making the platform more practical for individual investors.

The system follows a three-tier architecture consisting of a React frontend, a Node.js/Express backend, and a Python/FastAPI analytical service. Historical stock prices are collected using yfinance, while financial news is retrieved through the Finnhub API. Data preprocessing includes cleaning, normalization, and calculation of technical indicators such as moving averages, RSI, Bollinger Bands, ATR, and On-Balance Volume. The XGBoost model is trained using chronological data splits to avoid look-ahead bias, while FinBERT produces a recency-weighted sentiment index classified as bullish, neutral, or bearish. The RAG module retrieves similar historical cases and indicator descriptions from a FAISS vector database to generate concise explanations that clearly communicate model reasoning without providing investment advice.

Extensive testing demonstrates that TradeSense delivers reliable short-term predictions with response times suitable for interactive use while maintaining software stability through unit, integration, and functional testing. Case studies indicate that combining prediction probabilities, sentiment analysis, and explainable AI significantly improves user understanding compared to traditional stock analysis tools. However, the prototype has limitations, including reliance on free data sources, support for only single-user deployment, basic portfolio risk metrics, and reduced predictive performance during highly volatile market conditions. Future enhancements include incorporating macroeconomic and fundamental indicators, hybrid deep learning models, richer RAG knowledge bases using earnings reports and regulatory filings, scalable cloud deployment, and comprehensive user studies to further improve interpretability and real-world performance.

Conclusion

This paper presented TradeSense, a web-based decision support system aimed at making AI-driven market analytics available to individual investors. The system combines three components that are well-established in the research literature but rarely brought together in a single end-user product: an XGBoost classifier for short-term price trend prediction [1], a FinBERT-based news sentiment module [3], and a retrieval-augmented explanation engine [4] that transforms numerical outputs into plain-language summaries based on historical context. The prototype shows that integration itself adds value. Presenting prediction, sentiment, and explanation together makes each component more useful than any one of them would be alone. The divergence between a technical signal and a sentiment reading, for instance, is only visible when both are displayed at the same time, and it is exactly this kind of contextual tension that can lead to more thoughtful decision-making. We have been clear about the system’s limitations. TradeSense is a decision aid, not a trading system. Its accuracy is limited by the quality and coverage of its input data, and its predictions should be seen as one input among many rather than as definitive signals. The value it provides is not predictive alpha but interpretive clarity. It gives retail investors a more structured, evidence-based way to consider a position than what a price chart and a news feed can offer alone. Building on this foundation to create a more capable, production-ready platform is a natural and achievable next step.

References

[1] T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” in Proc. 22nd ACM SIGKDD Intl. Conf. on Knowledge Discovery and Data Mining, 2016, pp. 785–794. [2] S. M. Lundberg and S.-I. Lee, “A Unified Approach to Interpreting Model Predictions,” in Advances in Neural Information Processing Systems (NeurIPS), 2017, vol. 30. [3] D. Araci, “FinBERT: Financial Sentiment Analysis with Pre-trained Language Models,” arXiv preprint arXiv:1908.10063, 2019. [4] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” in Advances in Neural Information Processing Systems (NeurIPS), 2020, vol. 33, pp. 9459–9474. [5] FinRAG System, “Fin-RAG: A Retrieval-Augmented Generation System for Financial Documents,” Technical Report, 2024. [6] T. Sharma and S. Prasad, “Forecasting Stock Market Volatility Using XGBoost: A Time Series Analysis,” in Proc. IEEE Intl. Conf. on Emerging Technologies, 2022. [7] S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997. [8] K. Pardeshi et al., “Stock Market Price Prediction: A Hybrid LSTM and Sequential Self-Attention Architecture,” in Proc. IEEE ICDACS, 2023. [9] J. Devlin et al., “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” in Proc. NAACL, 2019, pp. 4171–4186. [10] A. Sinha, S. Kedas, and R. Kumar, “SEntFiN 1.0: Entity-Aware Sentiment Analysis for Financial News,” ACM Trans. Management Information Systems, vol. 14, no. 2, 2023. [11] M. Xiao et al., “Retrieval-Augmented Large Language Models for Financial Time Series Forecasting,” arXiv preprint arXiv:2502.05878, 2025.

Copyright

Copyright © 2026 A. Pavani, S. Bhavesh, Y. Sumanth, Dr. Bharathi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.