AMIT'S MODEL V1 // XGBOOST + RF + ELO + SQUAD + CAPTAINCY
RCB WIN PROBABILITY
40.9%
XGBoost 30.7% · RF 48.2% TOSS: NOT DECIDED
SRH WIN PROBABILITY
59.1%
Elo gap: +32 RCB · Squad adj: +3.2% TOSS: NOT DECIDED
HOW WE GOT TO 40.9%
BASE ENSEMBLE (XGBoost 60% + RF 40%)RCB 37.7%
SQUAD STRENGTH (RCB 5.71 vs SRH 4.10)+3.2%
CAPTAINCY (Patidar +0.5 · Kishan -0.5)INCLUDED ↑
WEATHER (March Bengaluru — dry season)0%
TOSS IMPACTNOT DECIDED
AMIT'S V1 FINALRCB 40.9%
MODEL PERFORMANCE // AMIT'S V1
XGBOOST (PRIMARY)
52.7%
5-FOLD CV ± 1.8% 1,090 MATCHES · 16 FEATURES · ELO
RANDOM FOREST
51.0%
5-FOLD CV ± 2.1% ENSEMBLE WEIGHT: 40%
TOP FEATURES // AMIT'S XGBOOST V1
Elo Diff
~15%
Team Form (t1/t2)
~13%
Venue Win Rate
~11%
H2H Record
~9%
Toss Impact
~5%
ACCURACYXGBoost 52.7% beats our GB 49.0%. Elo ratings are the key new feature — captures team momentum better than raw win rate.
MODEL GAPAmit's model: RCB 40.9% · Claude model: RCB 24.3% · Gap = 16.6%. Main reasons: H2H data (50% vs 40%), Elo captures RCB strength, squad score adds +3.2%.
HONEST TAKEBalanced estimate RCB 32-33%. Both models agree SRH are historical favourites. RCB edge = Kohli 9.6 form + home ground + Patidar confidence.
WHY SRH 75.7%?SRH scored 197 avg/match vs RCB 177. H2H 60-40 in SRH favour. Hazlewood OUT hurts RCB more than Cummins OUT hurts SRH.
VS AMIT'S MODELAmit's model: RCB 40.9% · Claude's model: RCB 24.3% · Gap = 16.6%. Amit's Elo captures RCB strength, H2H data differs (50% vs 40%).
HONEST TAKEBoth models agree SRH are historical favourites. True probability likely RCB 30–35%. Kohli 9.6 + home ground + Patidar confidence are the X-factors.
NOT LIVE
MATCH NOT STARTED
NEXT REFRESH: 5:00
RCB
—
BATTING
SRH
—
BOWLING
INITIALISING LIVE FEED...
Match: RCB vs SRH
Date: 28 MAR 2026 · 19:30 IST
Venue: M Chinnaswamy Stadium, Bengaluru
Auto-refreshes every 5 minutes.
› Predicted par score: 173 runs · Chinnaswamy chase win rate: 52% · Auto-refreshes every 5 mins
MATCH PREDICTION ENGINE · RCB vs SRH · 28 MARCH 2026
⚡ BUILT BY AMIT PANDIT · DATA SCIENTIST · LULU RETAIL
278K
Deliveries Analysed
1,169
IPL Matches
52.7%
Model Accuracy
16.6%
Model Disagreement
🗄️
DATA SOURCES
Kaggle IPL Dataset — 1,090 matches (2008–2024), deliveries + match results Cricsheet IPL JSON — 1,169 matches including IPL 2025, 278,205 ball-by-ball deliveries Cricsheet T20 JSON — 3,220 international T20 matches up to Mar 2026 CricketData.org API — live scores during match window (19:00–00:00 IST)
🏏
PLAYER FORM MODEL
Blended score 1–10 per player across 5 recent innings IPL 2024/25 (60%) + T20 Intl last 6 months (40%)
Weighted recency: last 3 innings carry double weight
Injury adjustments applied on top of base model score
Kohli 9.6 · Padikkal 8.4 · Unadkat 8.9
📊
FEATURE ENGINEERING
Elo ratings from 2022 (K=32, base 1500) — KKR 1579 · RCB 1512 · SRH 1479
Team form last 8 matches, venue win rate, H2H last 10
Batting/bowling form averages and form ratios
Toss decision adjustment (±3% at Chinnaswamy, 52% chase rate)
Squad strength score with captaincy bonus/penalty
🔬
INJURY ADJUSTMENTS
Applied on top of base ML probabilities per match Hazlewood OUT (RCB) → −5.0% for RCB Cummins OUT (SRH) → +4.0% for RCB Malinga DOUBTFUL (SRH) → +1.5% expected value
Net injury effect: +0.5% towards RCB
THE TWO MODELS
🤖 AMIT'S MODEL
Algorithm: XGBoost + Random Forest Ensemble Training: Kaggle IPL 2008–2024 (1,090 matches) Features: 16 — Elo, form, venue, H2H, toss, season CV Accuracy: 52.7% ± 1.8% (5-fold) Ensemble: 60% XGBoost + 40% RF Prediction: RCB 40.9% · SRH 59.1% Built in: Google Colab with scikit-learn + XGBoost
🧠 CLAUDE'S MODEL
Algorithm: Gradient Boosting + Random Forest Ensemble Training: Cricsheet IPL + T20 (1,146 matches, WC excluded) Features: 20 — adds batting/bowling form ratios CV Accuracy: 52.4% ± 4.7% (5-fold) Ensemble: 60% GB + 40% RF + injury layer Prediction: RCB 24.3% · SRH 75.7% Gap vs Amit: 16.6% — driven by H2H data difference