Kshitiz Negi

IPL Crunch '26: End-to-End Analytics Pipeline & Hypothesis Testing Engine

Analyse real IPL ball-by-ball match data to uncover meaningful patterns, answer cricket-related analytical questions, generate data-backed insights, and communicate findings through visualizations, analysis, and storytelling. Process My analysis began by challenging common cricket assumptions such as “winning the toss guarantees an advantage” and “matches are decided only in the death overs.” Instead of relying on opinions, we treated these beliefs as testable hypotheses using IPL ball-by-ball data. I designed a modular, config-driven Python pipeline rather than a single notebook to ensure reproducibility, maintainability, and cleaner separation between data processing and visualization. Initially, parsing 1,239 raw JSON files on every run took nearly 45 seconds, slowing experimentation and debugging. To solve this, I implemented a PyArrow Parquet caching layer, reducing load time to under one second. To avoid cherry-picked conclusions, we systematically tested eight hypotheses so insights were evidence-backed. Results We analysed 1,239 matches (~300K balls) and generated 12 premium charts. Toss winners win 51.6% of matches, which is statistically insignificant (p=0.263). I chose to report this null result because it refutes the quote, "toss is 50% of the game." Middle overs showed the largest winner/loser gap (Δ=7.8 runs) vs death (Δ=4.1). I chose to emphasize this because it proves games are won in overs 7–15. Shubman Gill leads batters (3,065 runs) and YS Chahal leads bowlers (99 wickets) for 2022–2026. Posting 200+ scores only wins 74.9% of matches (p<0.0001), debunking the "200 is a safe total" myth. Reflection First, I would invest in a schema-validation layer during loading. I initially ran the player statistics engine on the entire historical dataset, returning incorrect all-time leaders. I chose to pivot and hardcode a season list in config because it resolved the 5-season constraint (2022–2026) immediately, but a schema check would have caught it earlier. Second, I would build an interactive Plotly dashboard. I chose static matplotlib charts because they ensure exact print-layout control, but it limits real-time exploration. Finally, I would expand our hypothesis generator from 8 tests to 30.