MLB Performance Analytics applies Random Forest machine learning classifiers to MLB Statcast data to systematically classify 911 professional batters across three distinct offensive dimensions: overall performance, elite status, and plate discipline. By combining traditional batting statistics with advanced Statcast metrics — including exit velocity, launch angle, bat speed, and swing behavior — the project demonstrates how machine learning can uncover complex patterns that differentiate offensive profiles far beyond what conventional statistics reveal.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/Stronauta/MLB-Performance-Analytics/llms.txt
Use this file to discover all available pages before exploring further.
Classification Targets
Three target variables were engineered directly from the dataset, each capturing a different facet of batter performance:Overall Offensive Performance
Rendimiento_labels — Batters are divided into Low, Medium, and High performance tiers using pd.qcut on wOBA (Weighted On-Base Average), ensuring a balanced class distribution across all three groups.Elite Status
elite_hitter — A binary label that flags batters in the top 20% of wOBA as elite (1) and all others as non-elite (0). The model uses class_weight='balanced' to handle the inherent class imbalance.Plate Discipline
clase_disciplina_home — Batters are classified as Low (Baja), Medium (Media), or High (Alta) discipline using a composite score that weighs walk rate, pitch-taking behavior, plate appearances, and strikeout and swing-and-miss rates.Overall Offensive Performance (Rendimiento_labels)
The first classification target uses wOBA — a single statistic that weights each offensive outcome (singles, doubles, home runs, walks) by its true run value — as the foundation for grouping batters. Applying pd.qcut with three quantile bins guarantees that classes are approximately equal in size, avoiding the imbalanced training sets that can skew model accuracy. The resulting labels are Low, Medium, and High offensive performance.
Elite Status (elite_hitter)
The second target identifies batters who are genuinely exceptional rather than merely above average. Any batter whose wOBA falls at or above the 80th percentile of the full dataset is labeled elite (1); everyone else is labeled non-elite (0). Because the top 20% threshold produces a naturally unbalanced binary target, the Random Forest classifier for this model uses class_weight='balanced' to compensate.
Plate Discipline (clase_disciplina_home)
The third target measures how well a batter makes decisions at the plate. A composite disciplina_en_home score is computed as a weighted sum of five Statcast signals:
bb_percent(walk rate) — positive contribution (×0.30)takes(pitches taken without swinging) — positive contribution (×0.20)pa(plate appearances) — positive contribution (×0.10)k_percent(strikeout rate) — negative contribution (×0.20)swing_miss_percent(whiff rate) — negative contribution (×0.10)
pd.cut.
Dataset
The dataset covers 911 MLB batters drawn from the 2023–2025 regular seasons. Each row represents an aggregated batter profile containing 76 Statcast features that span traditional counting stats, batted-ball metrics, pitch-tracking data, and expected statistics.| Attribute | Value |
|---|---|
| Rows (batters) | 911 |
| Features | 76 |
| Seasons | 2023, 2024, 2025 (regular season) |
| Source | Baseball Savant / MLB Statcast |
woba, xwoba, ba, slg, obp, iso, babip, launch_speed, launch_angle, bat_speed, swing_length, barrels_total, barrels_per_bbe_percent, hardhit_percent, bb_percent, k_percent, swing_miss_percent, whiffs, swings, takes, and more.
Technology Stack
| Tool | Role |
|---|---|
| Python 3.9+ | Primary development language |
| Pandas | Data ingestion, cleaning, feature engineering, and target construction |
| NumPy | Numerical operations and array handling |
| Matplotlib / Seaborn | Exploratory data analysis and feature importance visualizations |
| Scikit-Learn | RandomForestClassifier, train_test_split, evaluation metrics (accuracy_score, precision_score, recall_score, f1_score) |
Project Structure
MLBStats.ipynb contains the complete end-to-end pipeline: data loading, exploratory analysis, feature engineering, target variable construction, model training for all three classifiers, performance evaluation, and feature importance visualization.
This project was developed as an academic capstone for Módulo 6 of the Data Science & Machine Learning program by Group 3 — D.A.T.A (Arlene Miniel, Elvis Rafael Rosado, Felix Mendoza, and Samir Ernesto Castillo). It is intended to demonstrate how machine learning models identify patterns in baseball statistics to differentiate offensive batter profiles.