Stock Recommendation System

This project presents an advanced Stock Recommendation System that integrates Sentiment Analysis of financial news with Technical Indicators to provide short-term stock recommendations. Additionally, it utilizes Long Short-Term Memory (LSTM) networks to forecast long-term stock price movements.

---

📌 Project Overview

Investors often struggle with making informed decisions due to the overwhelming volume of financial data and market sentiment. This project aims to enhance stock prediction by combining:

• 📈 Technical Indicators: Analyzing historical stock data trends and momentum indicators.
• 📰 Sentiment Analysis: Assessing market sentiment from financial news using Natural Language Processing (NLP).
• 📊 Price Prediction Model: Using LSTM networks to forecast future stock prices based on historical data.

🔍 Key Features:

✔ Multi-Model Approach: Combines Sentiment Analysis, Technical Indicators, and LSTM models.
✔ Short-Term & Long-Term Analysis: Provides both immediate trading signals and future price predictions.
✔ Real-Time Data Integration: Fetches stock market data using YFinance API and news sentiment from GNews API.
✔ Machine Learning-Powered Analysis: Utilizes SVM, Random Forest, and Multinomial Naïve Bayes for sentiment classification.

---

🏗️ Project Architecture

The system consists of three primary models:

1. 📉 Long-Term Price Prediction Model (LSTM)

   • Forecasts future stock prices using historical stock data.
   • Uses Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R² for evaluation.

2. 📊 Short-Term Recommendation Model

   • Integrates Technical Indicators and News Sentiment Model to provide Buy, Sell, or Hold recommendations.

3. 📡 Data Sources:

   • Stock Data: Extracted using YFinance API.
   • News Data: Fetched via GNews API.
   • Sentiment Dataset: FinancialPhraseBank dataset for training Sentiment Analysis model.

---

📊 Data Collection & Processing

Data Source	Details
YFinance API	Fetches 5 years of historical stock data (Open, Close, Volume, etc.)
GNews API	Retrieves financial news articles for sentiment analysis
Financial PhraseBank	Labeled financial statements for training Sentiment Analysis model

📌 Data Preprocessing:

✔ Removing duplicates and irrelevant columns.
✔ Normalizing stock price data using MinMaxScaler.
✔ Tokenizing & vectorizing text for sentiment analysis.

---

📈 Methodology

1️⃣ Long-Term Stock Price Prediction (LSTM)

• Uses four LSTM layers followed by Dropout layers to prevent overfitting.
• Optimizer: Adam, Loss Function: MSE, Epochs: 50, Batch Size: 32.
• Trained on 70% of the dataset, tested on 30%.
• Performance evaluation using MSE, RMSE, MAE, and R² metrics.

2️⃣ Short-Term Recommendation Model

(a) News Sentiment Analysis Model

• Classifies financial news articles into Positive, Negative, or Neutral using:
  ✔ Multinomial Naïve Bayes (MNB)
  ✔ Support Vector Machine (SVM) - Best Model
  ✔ Random Forest (RF)
• Best performing model: SVM with 79.75% accuracy.

(b) Technical Indicators Model

• Uses four key indicators:
  ✔ Moving Average (MA): Identifies trends.
  ✔ Relative Strength Index (RSI): Detects overbought/oversold conditions.
  ✔ Average Directional Index (ADX): Measures trend strength.
  ✔ On-Balance Volume (OBV): Analyzes volume flow.

✔ Buy Signal: If RSI < 30, Price > 50-day MA.
✔ Sell Signal: If RSI > 70, Price < 50-day MA.

3️⃣ Model Integration

• Final Recommendation Score = 40% Sentiment Analysis + 60% Technical Indicators
• Weighted combination determines Buy, Sell, or Hold signal.

---

🏆 Results & Accuracy

Model	Accuracy
LSTM Stock Price Prediction	RMSE: 4.2
Technical Indicators Model (TIM)	67.07%
News Sentiment Model (SVM)	79.75%
Final Weighted Model (TIM + Sentiment)	71.95%

🔹 The final Short-Term Model improved accuracy by 5% over the Technical Indicators alone.
🔹 The Hold signal accuracy significantly improved after integrating sentiment analysis.

---