readme_backtest

Cash 3 Lottery Historical Backtesting Simulation

This repo contains a detailed simulation engine that retroactively tests the accuracy of different predictive models on Georgia Cash 3 draws. Instead of predicting future numbers, this system walks day-by-day through historical data, pretending it's in the past and generating predictions using only past results.

---

🎯 Goal

To measure the true real-world effectiveness of various prediction models by:

• Running them on only the data available before each draw.
• Comparing their output to the known actual winning number. d
• Recording match statistics over hundreds of draws to determine if any method performs better than chance.

---

🔄 How It Works

1. Load full Cash 3 dataset (sorted from oldest to newest).
2. Starting after the first 10 draws (to build initial model memory):
   • At draw i:
     • Use draws 0 to i to generate predictions.
     • Compare prediction against draw i+1 (which the model has not seen).
3. Repeat for all draws until the latest.
4. Log the following:
   • Actual winning number
   • Each method's prediction
   • Hit type (Exact, 2-digit, 1-digit match)
   • Rolling accuracy for each method

---

🤖 Predictive Models Tested

• Markov Chain
• Frequency-Weighted
• ARIMA Time Series
• Hot/Cold Digit Analysis
• Sum & Parity Heuristics
• Digit Difference Patterns
• Gap Time Since Last Seen
• Modulo Cycles
• Entropy Scoring
• Machine Learning Model (RandomForest)
• Bayesian Probabilities

---

📈 Output

• historical_predictions_2025.csv — Daily model predictions + actuals
• historical_accuracy_2025.csv — Summary of how well each method performed

Example accuracy metrics logged:

• Exact Match %
• 2-digit Match %
• 1-digit Match %
• Top N performance if we pick top 10 guesses per day

---

🔍 Why This Is Valuable

✅ No need to wait for new draws to validate your models
✅ Shows how each method behaves over time — helps fine-tune or eliminate underperformers
✅ Let’s you train hybrid models or ensemble methods using the best of all techniques
✅ Allows for future "live" model deployment with high confidence in baseline performance

---

📁 Files

• full_backtest.py — Runs the full day-by-day simulation
• historical_predictions_2025.csv — All model outputs per day
• historical_accuracy_2025.csv — Final scorecard

---

---

**Built with curiosity by hamza/knightinfected ** """