📈 ArthaYukti – A Deep Learning-Driven Financial Sentiment Analysis & Stock Forecasting Platform

Built for INDOvateAI Sprint 2025 | Secured Second Prize 🏆

This project integrates FinBERT-based sentiment analysis with an LSTM-based stock price prediction model to provide a comprehensive market analysis. It dynamically assigns weightage to sentiment and price forecasts to improve investment decision-making.

💻 Source Code

You can check out our GitHub Repository for codes and models.

---

🚨 Problem Statement

📉 Investors face an overwhelming volume of real-time data, leading to delayed decisions and missed opportunities.
❌ Extracting accurate sentiment from unstructured sources is complex and error-prone, posing high-stakes risks.

---

🛠️ Solution Approach

✅ Custom Nifty50 database (2014–2025, 129,377 rows) → Cleaned & preprocessed for time-series forecasting.
✅ FinBERT-based Sentiment Extraction → Trained on 1.4M financial headlines to extract bullish, bearish, or neutral sentiment.
✅ LSTM-based Time-Series Prediction → Forecasts stock price trends based on historical market data.
✅ User-Friendly Dashboard → Displays prediction charts, source citations, analytics, and investment recommendations.

---

🏗️ System Architecture

1️⃣ Data Acquisition & Reliability

• Customizable ETL from yfinance for accurate real-time data.
• Fully documented & version-controlled codebase on GitHub.

2️⃣ NLP & Sentiment Analysis

• Structured LLMs for multi-language financial news processing.
• Sentiment classification (bearish, bullish, neutral) via FinBERT.

3️⃣ Forecasting & Dynamic Analysis

• LSTM-based stock price forecasting.
• Weighted analysis combining sentiment & confidence scores using a custom formula.
• Provides actionable insights for investors.

4️⃣ Real-Time Processing & Scalability

• Low latency real-time input processing.
• Extensive Flask endpoints for API-driven predictions.
• Cache-based state management for multi-user support.

5️⃣ Visualization & User Empowerment

• Multiple interactive graph view options.
• Source verification for user confidence & validation.

---

🎥 Demo

Demo Video

---

🚀 Features

✅ LSTM & FinBERT Integration – Combines deep learning & NLP for robust stock forecasting.
✅ Sentiment Analysis from Financial News – Extracts real-time news sentiment.
✅ Custom Dynamic Weight Assignment – Adjusts importance of sentiment vs. prediction confidence.
✅ Real-Time Market Predictions – Generates buy/hold/sell signals.
✅ Market Analysis Dashboard – Displays real-time sentiment, trend predictions, and historical analysis.
✅ Multi-Market Adaptability – Can be extended to crypto, forex, and commodities.
✅ Research & Analytics Tool – Useful for financial researchers & institutions.