Sika: A Trend Prediction System

A local-first CLI tool for financial market analysis and trend prediction using advanced technical indicators

---

🎯 Overview

Sika is a machine learning system designed to predict foreign exchange and commodity market trends using neural networks and technical indicators. Built as a local-first CLI application, it provides a clean, intuitive interface for training models, making predictions, and tracking metrics.

---

✨ Features

Technical Features

• Feature Engineering: TTM Trend, MACD, RSI, ADX, Stochastic RSI, and INC/DEC
• Model Architecture: Multi-layer perceptron
• Data Processing: Min-Max Scaling
• Model Persistence: Serialized models and scalers as .pkl files
• Comprehensive Logging: Logging with detailed execution traces

---

📊 Project Metrics

Performance Characteristics

Metric	Value
Model Type	Multi-Layer Perceptron Classifier
Training Algorithm	L-BFGS Optimizer
Default Iterations	10,000
Feature Count	7 technical indicators
Train/Test Split	80/20
Typical Accuracy	80-90% (market-dependent)

Stack Overview

Python 3.12.9
├── Data Processing
│   ├── pandas (2.3.3)
│   └── numpy (2.3.4)
├── Machine Learning
│   ├── scikit-learn (1.7.2)
│   ├── scipy (1.17.1)
│   └── ta-lib (0.6.8)
└── CLI & Display
    ├── rich (14.2.0)
    └── colorama (0.4.6)

---

🚀 Quick Start

Prerequisites

• Python 3.12.9
• uv package manager

Installation

# Clone or download the repository
git clone <repository-url>
cd sika

# Create virtual environment
uv sync

# Install dependencies
uv pip install -r requirements.txt

# Create configuration file
cp .env.example .env

Your First Prediction

# Start interactive mode (recommended)
uv run main.py

# Or run directly
uv run main.py --mode predict --pair XAUUSD --open 2650.50

---

⚙️ Configuration

Environment Variables

Create a .env file in the project root:

# Data Directories
RAW_DATA_DIR=data/raw
PROCESSED_DATA_DIR=data/processed
MODEL_DIR=models
LOGS_DIR=logs

# Trading Pairs (comma-separated)
TRADING_PAIRS=XAUUSD,EURUSD,GBPUSD

# Data Configuration
START_DATE=2020-01-01
RANDOM_STATE=42

# Model Hyperparameters
TRAIN_SPLIT=0.8
ACTIVATION=logistic
SOLVER=lbfgs
LEARNING_RATE=adaptive
LEARNING_RATE_INIT=0.03
MAX_ITER=10000
MOMENTUM=0.2
EARLY_STOPPING=True

# Feature Selection
SELECTED_FEATURES=TTM_TRND_6,MACD_12_26_9,RSI_14,ADX_14,STOCHRSIk_10_14_3_3,INC_1,DEC_1

TIINGO_KEY = your_tiingo_api_key

---

🏗️ Architecture

System Pipeline

flowchart TD
    A["<b>data/raw/*.csv</b><br/>(OHLC CSV files)"]
    B["<b>Preprocessing & Feature Eng.</b><br/>scripts/data.py<br/>scripts/indicators.py<br/>(scaling, selected features)"]
    C["<b>Training (models)</b><br/>scripts/train.py"]
    D["<b>Config & CLI</b><br/>.env + config.py<br/>main.py (CLI entrypoint)"]
    E["<b>Prediction Engine</b><br/>scripts/predict.py<br/>loads models/PAIR_mlp_classifier.pkl"]
    F["<b>Output: CLI display & logs</b><br/>scripts/display.py & logs/sika.log"]

    A --> B
    B --> C
    B --> D
    D --> E
    E --> F

Loading

• High-level pipeline: raw data → preprocessing & indicators → train → model artifacts → prediction → CLI display & logs.

---

📁 Project Structure

sika/
├── main.py                 # CLI entry point
├── config.py              # Configuration management
├── pyproject.toml         # Project metadata
├── requirements.txt       # Python dependencies
│
├── scripts/               # Core modules
│   ├── __init__.py
│   ├── train.py          # Training pipeline
│   ├── predict.py        # Prediction engine
│   ├── data.py           # Data loading & preprocessing
│   ├── indicators.py     # Technical indicators
│   ├── log.py            # Accuracy logging
│   ├── logger.py         # File logging setup
│   └── display.py        # CLI display utilities
│
├── data/                 # Data storage
│   ├── raw/              # Original OHLC data (CSV)
│   └── processed/        # Processed features
│
├── models/               # Model persistence
│   ├── PAIR_mlp_classifier.pkl
│   ├── PAIR_scaler.pkl
│   └── PAIR_metadata.json
│
└── logs/                 # Execution logs
    └── sika.log

---

📈 Technical Indicators

Sika uses 7 strategically selected technical indicators:

1. TTM Trend (6-period)

• Measures trend strength
• Input: Last 6 candles
• Output: Scaled trend intensity

2. MACD (12, 26, 9)

• Momentum oscillator
• Parameters: Fast=12, Slow=26, Signal=9
• Captures trend changes and momentum

3. RSI (14-period)

• Relative Strength Index
• Range: 0-100 (Overbought/Oversold)
• Identifies reversal opportunities

4. ADX (14-period)

• Average Directional Index
• Measures trend strength
• Range: 0-100 (Strong/Weak)

5. Stochastic RSI

• RSI applied to RSI values
• Period: 10, Smoothing: 3
• Momentum confirmation

6. Price Increase Ratio (INC_1)

• Percentage of up candles
• 1-period lookback
• Recent bullish pressure

7. Price Decrease Ratio (DEC_1)

• Percentage of down candles
• 1-period lookback
• Recent bearish pressure

---

🔧 Customization

Adding New Trading Pairs

1. Update .env configuration:

   TRADING_PAIRS=XAUUSD,EURUSD,YOUR_NEW_PAIR

2. Train model:

   python main.py --mode train --pair YOUR_NEW_PAIR

Modifying Technical Indicators

Edit scripts/indicators.py to add or modify indicators:

def custom_indicator(df):
    """Add your custom indicator here"""
    return calculated_values

Then update SELECTED_FEATURES in .env.

Tuning Model Hyperparameters

Modify .env values:

# For more training: increase MAX_ITER
MAX_ITER=15000

# For faster convergence: adjust learning rate
LEARNING_RATE_INIT=0.05

# For stronger regularization: increase MOMENTUM
MOMENTUM=0.5

---

📊 Model Performance

Typical Characteristics

• Accuracy Range: 80-90% (market-dependent)
• Training Time: 30-120 seconds per pair
• Prediction Latency: <100ms
• Memory Footprint: 100-200MB per session
• Data Requirements: 3-5 years historical data

Note: Accuracy depends heavily on market conditions, pair volatility, and indicator stability. Regular retraining recommended when market regimes change.

---

🐛 Troubleshooting

Issue: "Model not found for pair"

# Solution: Train the model first
python main.py --mode train --pair XAUUSD

Issue: "Insufficient data"

# Solution: Ensure raw CSV exists in data/raw/
# Format should be: PAIRRAW.csv
# Example: XAUUSDRAW.csv

Issue: "Feature mismatch during prediction"

# Solution: Retrain model with current indicators
python main.py --mode train --pair XAUUSD

Issue: High memory usage

# Solution: Reduce MAX_ITER in .env
MAX_ITER=5000

Issue: Import errors

# Solution: Reinstall dependencies
pip install -r requirements.txt

---

🎯 Use Cases

Portfolio Analysis

Monitor multiple forex pairs and commodities with consistent ML-based analysis.

Strategy Development

Test trading ideas and validate signals against historical predictions.

Risk Analysis

Identify trend changes early with high-accuracy predictions.

Market Learning

Understand technical analysis and ML applications in finance.

Data Exploration

Analyze market patterns and indicator relationships.

---

🏆 Tips for Best Results

1. Use Quality Data: Ensure OHLC data is complete and accurate
2. Retrain Regularly: Models degrade over time as market conditions change
3. Verify Predictions: Log actual results to track accuracy
4. Test Thoroughly: Backtest strategies before live use
5. Monitor Accuracy: Track metrics over time to catch degradation
6. Optimize Hyperparameters: Experiment with different settings for your pairs
7. Handle Missing Data: Clean data before training

---

📚 Technical Stack

Component	Version	Purpose
Python	3.12.9	Runtime environment
scikit-learn	1.7.2	Machine learning
pandas	2.3.3	Data manipulation
numpy	2.3.4	Numerical computing
ta-lib	0.6.8	Technical analysis
rich	14.2.0	Terminal UI
joblib	1.5.3	Model serialization
python-dotenv	1.2.2	Configuration

---

📞 Support

For questions or issues: reachout at: newmankelvin14@gmail.com

---

⚠️ Disclaimer

Sika is provided for educational and research purposes only. Trading in financial markets carries substantial risk of loss. Past performance is not indicative of future results. Always conduct thorough backtesting and due diligence before using predictions for live trading. The authors are not responsible for trading losses or decisions made based on predictions from this system.

---

⭐ If you find this project useful, consider giving it a star!