🚀 BursaLens - Indonesian Stock Market Prediction & News Analysis

An advanced stock market analysis platform combining deep learning LSTM predictions with real-time news scraping. Built for the Indonesian market (IDX), featuring comprehensive technical analysis, fundamental metrics, and news integration.

✨ Features

🤖 LSTM Stock Prediction

• Advanced Neural Network Architecture

  • Multi-layer LSTM with dropout layers
  • Sequence-based prediction (30-days lookback)
  • Dynamic feature engineering
  • Robust scaling and preprocessing
  • Early stopping for optimal training

• Technical Indicators

  • Moving Averages (SMA20, SMA50)
  • Relative Strength Index (RSI)
  • MACD (Moving Average Convergence Divergence)
  • Bollinger Bands
  • Rate of Change (ROC)
  • Volume Analysis

• Fundamental Analysis

  • Market Capitalization
  • P/E Ratio
  • ROE (Return on Equity)
  • ROA (Return on Assets)
  • Current Ratio
  • Free Cash Flow Growth
  • Dividend Yield

• Model Performance Metrics

  • MAPE (Mean Absolute Percentage Error)
  • R-squared Score
  • Directional Accuracy
  • RMSE (Root Mean Square Error)
  • Threshold Accuracy

📰 News Integration

• Real-time news scraping from Liputan6
• Market sentiment analysis
• Article categorization
• Image and media extraction
• Timestamp standardization

🧠 LSTM Model Architecture

Data Preprocessing

def prepare_data(self, df):
    # Feature Engineering
    features = [
        'Close', 'Returns', 'SMA20', 'SMA50',
        'RSI', 'ROC', 'MACD', 'BB_width',
        'Volume_Ratio'
    ]
    
    # Robust Scaling
    scaled_data = self.scaler.fit_transform(data[features])
    
    # Sequence Creation
    X, y = [], []
    for i in range(self.lookback_period, len(scaled_data)):
        X.append(scaled_data[i-self.lookback_period:i])
        y.append(scaled_data[i, 0])

Model Structure

def create_model(self, input_shape):
    model = Sequential([
        LSTM(units=100, return_sequences=True, input_shape=input_shape),
        Dropout(0.2),
        LSTM(units=70, return_sequences=False),
        Dropout(0.1),
        Dense(units=50),
        Dense(units=1)
    ])
    
    model.compile(optimizer=Adam(learning_rate=0.001), loss='mse')
    return model

Technical Indicators

def add_features(self, df):
    # Price Features
    df['Returns'] = df['Close'].pct_change()
    df['SMA20'] = df['Close'].rolling(window=20).mean()
    df['SMA50'] = df['Close'].rolling(window=50).mean()

    # Momentum
    df['RSI'] = ta.momentum.rsi(df['Close'])
    df['ROC'] = ta.momentum.roc(df['Close'])

    # Trend
    df['MACD'] = ta.trend.macd_diff(df['Close'])

    # Volatility
    bollinger = ta.volatility.BollingerBands(df['Close'])
    df['BB_width'] = bollinger.bollinger_hband() - bollinger.bollinger_lband()

    # Volume
    df['Volume_SMA'] = df['Volume'].rolling(window=20).mean()
    df['Volume_Ratio'] = df['Volume'] / df['Volume_SMA']

Prediction Process

def predict_future(self, data, days=30):
    # Prepare recent data
    X = self.prepare_sequence(data)
    
    # Generate predictions
    predictions = []
    current_sequence = X[-1:]
    
    for _ in range(days):
        # Predict next value
        pred = self.model.predict(current_sequence)
        predictions.append(pred[0][0])
        
        # Update sequence for next prediction
        current_sequence = self.update_sequence(current_sequence, pred)
    
    return self.inverse_transform(predictions)

📊 Supported Stocks

Major Indonesian stocks including:

• BBCA (Bank Central Asia)
• BBRI (Bank Rakyat Indonesia)
• BMRI (Bank Mandiri)
• TLKM (Telkom Indonesia)
• ASII (Astra International)
• UNVR (Unilever Indonesia)
• And many more...

🌐 API Endpoints

Stock Prediction

Endpoint	Method	Description
/stock/<symbol>	GET	Get predictions and analysis

News API

Endpoint	Method	Description
/news	GET	Latest market news
/details	GET	Detailed article content

💻 Frontend Integration

Stock Prediction Component

import { LineChart, Line, XAxis, YAxis, Tooltip } from 'recharts';

const StockChart = ({ historicalData, predictions }) => {
  return (
    <div className="w-full h-96">
      <LineChart data={[...historicalData, ...predictions]}>
        <Line type="monotone" dataKey="price" stroke="#8884d8" />
        <Line type="monotone" dataKey="prediction" stroke="#82ca9d" />
        <XAxis dataKey="date" />
        <YAxis />
        <Tooltip />
      </LineChart>
    </div>
  );
};

🚀 Docker Deployment

docker-compose.yml

version: '3.8'
services:
  backend:
    build: ./backend
    ports:
      - "5000:5000"
    volumes:
      - ./backend:/app
    environment:
      - FLASK_ENV=production

📊 Model Performance

The LSTM model is evaluated using multiple metrics:

• MAPE: Typically 2-5% for short-term predictions
• Directional Accuracy: 40-50%
• R-squared Score: 77-84%

🛠️ Technical Requirements

Backend

• Python 3.8+
• TensorFlow 2.x
• Flask
• pandas
• numpy
• scikit-learn
• BeautifulSoup4
• yfinance

Frontend

• Node.js 16+
• Next.js 13+
• TypeScript
• Tailwind CSS
• Recharts
• Font Awesome

🚀 Getting Started

Local Development

1. Clone and Setup

git clone https://github.com/yourusername/bursalens.git
cd bursalens

2. Backend

cd backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
python app.py

3. Frontend

cd frontend
npm install
npm run dev

AWS EC2 Deployment

1. Instance Setup

sudo yum update -y
sudo yum install -y docker
sudo service docker start

2. Application Deployment

docker-compose up -d

🔄 Maintenance

Model Retraining

• Automated daily data collection
• Weekly model retraining
• Performance monitoring
• Hyperparameter optimization

News Updates

• Real-time scraping
• Hourly content refresh
• Daily archives

📝 License

MIT License - see LICENSE

👥 Authors

• Felix Mulya
• Email: [email protected]
• GitHub: @felixxmulya

🙏 Acknowledgments

• TensorFlow
• yfinance
• BeautifulSoup4
• Next.js
• Tailwind CSS
• Liputan6

---

⭐️ Star this repo if you find it helpful!