Hardik Mahindroo

IPL Crunch: 5-Season Macro Analysis & T20 Winning Framework

T20 cricket strategies are heavily dictated by subjective opinions and misleading raw aggregates. This project replaces conjecture with data analytics by processing a ball-by-ball dataset across five consecutive IPL seasons (2021–2025). The goal is to isolate true match drivers by evaluating the statistical impact of winning the toss, normalizing phase lengths to identify which over-block dictates victory, and identifying macro-window leaderboards for elite player performance. Process Parsed raw Cricsheet JSON match logs into structured pandas dataframes, filtering specifically for consecutive IPL seasons. Segmented delivery-level data into chronological over blocks and engineered a Runs Per Over (RPO) disparity feature to normalize unequal phase lengths. Advanced past basic data grouping to build an advanced conditional probability matrix, mapping how early success compounds into match-winning momentum. Aggregated macro-window player data over a five-season window to compute contextual metrics like Strike Rates and Economies. Exported crisp figures using Seaborn and packaged the final insights into an elite presentation layout in Canva. Results • Toss Myth Debunked: Proved that winning the toss offers a negligible 1.60% win-rate advantage. • Phase Intensity: Normalizing unequal over-blocks revealed that the Powerplay drives match outcomes through an elite 1.05 RPO separation. • Momentum Framework: Proved that winning the Powerplay unlocks an invincible 90.00% win rate if carried into the Middle Overs. Conversely, losing the Powerplay drops middle-over success to a mere 52.60% coin toss. Reflection • Integrate Ball Tracking: Incorporate hawk-eye trajectory metrics to analyze line, length, and optimal release points for the top 5 bowlers. • Incorporate Game-State Dynamics: Factor in real-time DLS (Duckworth-Lewis-Stern) pressure indexes and historical stadium boundaries to contextualize scoring acceleration. • Deploy Predictive Machine Learning: Move past static conditional probabilities to train a live XGBoost or LSTM model for real-time win-probability forecasting.