Decision Intelligence System for IPL Cricket
Not just analytics. An AI system that tells you what to do, why it matters, and how confident it is — on every screen.
Most cricket dashboards show data. CricketBrain AI makes decisions.
| Other platforms | CricketBrain AI |
|---|---|
| Shows win probability | Explains WHY and what to do next |
| Displays stats | Generates actionable decisions |
| Static charts | Live pressure + momentum tracking |
| Generic rankings | Clutch / Stable / High-Risk classification |
| One fantasy team | 3 teams — Safe / Grand League / Differential |
📁 IPL.csv (2008–2025, 278K+ deliveries)
│
▼
┌─────────────────────────────────────┐
│ ETL Pipeline │
│ • Auto-detects any column format │
│ • Team/venue normalization │
│ • Phase: Powerplay / Middle / Death│
│ • Parquet + CSV output │
└──────────────────┬──────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Feature Engineering (70+ features) │
│ • Zero data leakage (shift+rolling)│
│ • Rolling win rates (last 3/5/10) │
│ • H2H stats (time-aware) │
│ • Pressure Index + Momentum Index │
│ • EMA batting form (α=0.3) │
│ • xRuns, xWickets │
└──────────────────┬──────────────────┘
│
▼
┌─────────────────────────────────────┐
│ ML Pipeline │
│ • XGBoost (Optuna, 50 trials) │
│ • LightGBM (Optuna, 40 trials) │
│ • CatBoost (500 iterations) │
│ • Stacking Ensemble (LR meta) │
│ • Isotonic calibration │
│ • SHAP TreeExplainer │
└──────────────────┬──────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Intelligence Engines │
│ • Decision Engine (5K simulations) │
│ • Monte Carlo (10K simulations) │
│ • Fantasy ILP Optimizer (PuLP) │
│ • Weakness Detector │
│ • Insight Generator (NL) │
└──────────────────┬──────────────────┘
│
▼
┌─────────────────────────────────────┐
│ 24-Page Streamlit Dashboard │
│ + FastAPI Backend (15 endpoints) │
└─────────────────────────────────────┘
- Stacking Ensemble — XGBoost + LightGBM + CatBoost meta-learned by Logistic Regression
- Optuna Tuning — 50–90 trial Bayesian hyperparameter search per model
- SHAP Explainability — every prediction shows top feature contributions
- Calibrated Probabilities — isotonic regression, probabilities are reliable
- Zero Data Leakage — all rolling features use
shift(1)— model never sees the future - TimeSeriesSplit CV — 5-fold chronological validation
Every AI decision includes all five components:
- Data Evidence — exact RR deficit, pressure index shown numerically
- Simulation Support — 5,000 runs comparing Aggressive vs Safe strategies with win %
- Confidence Level — HIGH / MEDIUM / LOW with reason
- Multiple Options — Aggressive and Conservative with win % each
- Consequence — "If not followed → win probability drops by X%"
- Sensitivity Analysis — what if required RR rises? What if a wicket falls?
- Ball-by-ball win probability with per-over delta chart
- Turning Point Detection — highlights overs where WP shifted ≥8%
- Pressure Index — 0–100 based on required rate, wickets, overs left
- Momentum Index — last 12 balls, exponential decay weighted (−100 to +100)
- Required RR vs Current RR gap chart with colour zones
- 10,000 match simulations per prediction
- Win %, 95% confidence interval, score distributions (P10–P90)
- Dominant / contested / upset scenario breakdown
- Auto-generated natural language match summary
- 3 teams generated simultaneously — Safe, Grand League, Differential
- Ceiling / Floor / Variance per player shown in pool table
- Ownership Prediction — estimated % of teams picking each player
- WHY THIS PLAYER reasoning for every pick
- Pitch Scenario Mode — batting/bowling pitch adjusts fantasy weights
- ILP (PuLP CBC) with greedy fallback
- Alternative swap suggestions per team
- 4-way Classification — 🔥 Clutch / ⚡ High-Risk / 🛡️ Stable / 🎯 Flat-track Bully
- EMA Form Line — exponentially weighted recent form plotted on chart
- Consistency Score (0–100) + Volatility Score (0–10) as gauges
- Clutch Score — pressure performance vs normal performance ratio
- xRuns / xWickets — expected stats beyond simple averages
- Phase-level breakdown with above/below career baseline
- Pitch DNA Classification — Batting Paradise / Balanced / Bowler-Friendly / Seam Track
- Spin vs Pace Analysis — economy rate and wicket rate head-to-head
- Phase Dominance — which phase defines this ground
- AI Best XI — venue-specific player selection from historical data
- Strategy Advisor — data-backed bat first vs chase recommendation
- Playoff Probability — Monte Carlo qualification simulation per team
- Form Trend — last 5 matches as 🟢🔴 visual
- Expected Final Position column
- Full colour-coded points table
- Season-by-season accuracy chart
- Calibration curve (Reliability Diagram)
- All model metrics vs random baseline
- Full temporal validation methodology
| # | Page | Key Intelligence |
|---|---|---|
| 1 | 🏆 League Overview | Season trends, team dominance |
| 2 | 🔍 Player Profile | Stats, phase breakdown, AI weaknesses |
| 3 | ⚔️ Player Comparison | Side-by-side, radar chart |
| 4 | 🥊 Batsman vs Bowler | Matchup matrix, dismissal types |
| 5 | 🤝 Partnership Tracker | Best pairs, RPO analysis |
| 6 | ⚔️ Team Comparison | H2H record, cumulative wins |
| 7 | 🤖 AI Match Predictor | Ensemble + SHAP + Monte Carlo |
| 8 | 🎯 Toss Advisor | Venue data, bowling strategy |
| 9 | 📈 Win Probability | Ball-by-ball WP + turning points |
| 10 | 📋 Season Points Table | Standings + playoff probability |
| 11 | 🌡️ Run Heatmap | Phase analysis, anomaly detection |
| 12 | 📊 Form Tracker | HOT/DECLINING + risk score |
| 13 | ⚡ Breakout Players | Why breakout + sustainability |
| 14 | 💰 Fantasy Optimizer | ILP team + ownership strategy |
| 15 | 🏅 Player Rankings | Impact score + style clusters |
| 16 | 🏟️ Venue Analysis | Pitch type + player suitability |
| 17 | 🧬 Player Similarity | Role match + best replacement |
| 18 | 📖 About & Model Info | Architecture + metrics vs baseline |
| 19 | 🧠 Decision Engine | Full 5-component AI decision |
| 20 | ⚡ Pressure & Momentum | Pressure index + momentum graph |
| 21 | 🔬 Deep Player Analytics | Classification + xRuns + clutch |
| 22 | 💎 Elite Fantasy Teams | 3-team system + ceiling/floor |
| 23 | 🏟️ Venue DNA Advanced | Spin/pace + best XI + phase |
| 24 | 📊 Backtesting Dashboard | Accuracy + calibration curve |
cricketbrain/
├── etl/
│ ├── data_cleaning.py # IPL pipeline, auto-detects column formats
│ ├── feature_engine.py # 70+ features, zero leakage
│ └── insight_generator.py # NL insights, toss advisor
│
├── ml/
│ ├── train.py # Full ML pipeline with Optuna + SHAP
│ └── weakness_detector.py # Phase/matchup weakness analysis
│
├── simulation/
│ └── monte_carlo.py # 10,000-run match simulator
│
├── optimizer/
│ └── fantasy_optimizer.py # ILP Dream11 optimizer, 3 strategies
│
├── api/
│ └── main.py # FastAPI, 15 endpoints, Pydantic v2
│
├── app/
│ ├── app.py # Main router, 24-page sidebar nav
│ ├── upgraded_pages.py # Pages 9–18 with AI intelligence
│ ├── phase2_pages.py # Pages 19–24, Decision Engine
│ ├── decision_engine.py # Pressure/Momentum/xRuns/Decisions
│ └── intelligence.py # Form classification, risk scoring
│
├── data/
│ ├── raw/IPL.csv # ← Place Kaggle download here
│ ├── cleaned/ # matches.csv + deliveries.csv
│ ├── features/ # Engineered feature tables
│ ├── models/ # Trained PKLs + metrics.json
│ └── shap/ # SHAP explainers + importance CSVs
│
├── init_db.py # SQLite schema setup
├── requirements.txt
├── Dockerfile
└── docker-compose.yml
Download from Kaggle and place at data/raw/IPL.csv:
https://www.kaggle.com/datasets/chaitu20/ipl-dataset2008-2025
pip install -r requirements.txt
# Mac M-series only
brew install libomppython etl/data_cleaning.py
python etl/feature_engine.py
python init_db.pypython ml/train.py
# ~5–15 minutes | Trains XGBoost → LightGBM → CatBoost → Stacking → Calibrationstreamlit run app/app.py
# http://localhost:8501uvicorn api.main:app --host 0.0.0.0 --port 8000 --reload
# Docs at http://localhost:8000/docsdocker-compose up --build
# Dashboard → http://localhost:8501
# API → http://localhost:8000/docs| Endpoint | Method | Description |
|---|---|---|
/predict |
POST | Match prediction + SHAP + simulation |
/simulate |
POST | Monte Carlo simulation |
/fantasy/optimize |
POST | Dream11 ILP optimizer |
/players/{name}/stats |
GET | Career stats |
/players/{name}/weakness |
GET | Weakness analysis |
/players/{name}/insights |
GET | AI form insights |
/matchup |
POST | Batsman vs bowler H2H |
/teams/{name}/insights |
GET | Team insights + strategy |
/match/preview |
GET | Pre-match AI briefing |
/shap/{model} |
GET | SHAP feature importance |
/model/info |
GET | Metrics + feature list |
/health |
GET | System health |
| Innovation | Details |
|---|---|
| Pressure Index | Proprietary: f(required_rate, wickets_left, overs_left) |
| Momentum Index | Last 12 balls, exponential decay weighted |
| EMA Form | α=0.3 exponentially weighted batting average |
| Zero Leakage | All features use shift(1).rolling() |
| Clutch Score | High-pressure vs normal performance ratio |
| Pitch DNA | Auto pitch classification from venue stats |
| ILP Optimizer | PuLP CBC — guaranteed optimal team, not greedy |
| xRuns/xWickets | Expected value metrics beyond raw averages |
| Calibration | Isotonic regression — probabilities are meaningful |
| Backtesting | Season-by-season accuracy with calibration curve |
| Layer | Technology |
|---|---|
| Data | Pandas, NumPy, PyArrow |
| ML | XGBoost, LightGBM, CatBoost, scikit-learn |
| Tuning | Optuna (Bayesian) |
| Explainability | SHAP |
| Optimization | PuLP (ILP) |
| Simulation | NumPy Monte Carlo |
| API | FastAPI + Uvicorn |
| Dashboard | Streamlit + Plotly |
| Database | SQLite |
| Deployment | Docker + docker-compose |
MIT License — free for personal, educational, and portfolio use. 📊 Dashboard Preview
🔥 IPL League Overview
👨💻 Author
Pradip Kumar Verma
AI & Data Science StudentPassionate about building real-world ML systems 🚀