Classification Targets: Three Ways to Measure Batter Performance

Three classification labels are constructed directly from the raw Statcast data. Rather than predicting a numeric value, each target frames batter performance as a classification problem — grouping players into discrete, meaningful categories based on real offensive metrics. This approach allows the Random Forest models to learn which statistical signatures distinguish performance tiers, elite ability, and plate decision-making quality from one another.

Overall Performance
Elite Status
Plate Discipline

Rendimiento_labels — Overall Performance

The first target, Rendimiento_labels, captures a batter’s general offensive production by binning their wOBA (Weighted On-Base Average) into three equal quantile groups: Bajo (Low), Medio (Medium), and Alto (High).

Why wOBA?

wOBA is a comprehensive offensive metric that assigns different weights to each type of hit based on its actual run value — a walk is worth less than a single, which is worth less than a double, and so on. This makes it a far more accurate measure of overall offensive contribution than simpler stats like batting average or on-base percentage.

Label Construction

Using pd.qcut with q=3 ensures that each class receives approximately the same number of players, producing balanced training labels:

df["Rendimiento_labels"] = pd.qcut(
    df["woba"],
    q=3,
    labels=["Bajo", "Medio", "Alto"]
)

Classes

Label	Meaning	Quantile Range
`Bajo`	Low performance	Bottom third of wOBA
`Medio`	Medium performance	Middle third of wOBA
`Alto`	High performance	Top third of wOBA

Quantile-based splitting guarantees balanced class sizes by design. This avoids the class imbalance issues that can arise when using fixed thresholds on skewed distributions.

elite_hitter — Elite Status

The second target, elite_hitter, asks a binary question: is this batter among the truly exceptional offensive performers in the league? A player is labeled elite (1) if their wOBA falls in the top 20% of all batters; otherwise they are labeled non-elite (0).

Why Binary?

While Rendimiento_labels partitions performance into thirds, the elite model focuses specifically on separating the highest-impact batters from the rest of the field. This framing is useful for scouting and roster construction contexts where identifying peak performers matters more than ranking the entire population.

Label Construction

df["elite_hitter"] = (df["woba"] >= df["woba"].quantile(0.80)).astype(int)

Class Distribution

Out of 857 batters in the cleaned dataset:

Class	Label	Count
Non-Elite	`0`	684
Elite	`1`	173

This 4:1 class imbalance is intentional — the elite tier is, by definition, rare. The Random Forest model for this target uses class_weight='balanced' and stratified splitting to handle this imbalance during training.

The 80th-percentile threshold was chosen to mirror real-world usage: roughly the top 20% of MLB batters in any given season are considered genuinely elite offensive contributors.

plate_discipline — Plate Decision Quality

The third target measures something different from raw production: the quality of a batter’s decision-making at the plate. Rather than measuring how often a batter gets hits, it evaluates whether they correctly identify pitches in and out of the strike zone and respond accordingly.

Composite Score

A weighted composite score (disciplina_en_home) is first computed from discipline-related Statcast metrics, rewarding patience and penalizing poor contact decisions:

df["disciplina_en_home"] = (
    df["bb_percent"]        * 0.30  # Walk rate — rewards patience
    + df["takes"]           * 0.20  # Pitches taken — rewards selectivity
    + df["pa"]              * 0.10  # Plate appearances — activity proxy
    - df["k_percent"]       * 0.20  # Strikeout rate — penalizes poor decisions
    - df["swing_miss_percent"] * 0.10  # Swing-and-miss rate — penalizes bad contact
    - df["whiffs"]          * 0.05  # Raw whiff count
    - df["swings"]          * 0.05  # Raw swing count
)

Players are then classified into three tiers using fixed score bins:

df["clase_disciplina_home"] = pd.cut(
    df["disciplina_en_home"],
    bins=[-999, 200, 800, 1200],
    labels=["Baja", "Media", "Alta"]
)

Classes

Label	Meaning
`Baja`	Low plate discipline
`Media`	Average plate discipline
`Alta`	High plate discipline

Key Input Metrics

Metric	Role
`bb_percent`	Walk rate — primary positive indicator
`k_percent`	Strikeout rate — primary negative indicator
`swing_miss_percent`	Contact failure rate
`takes`	Pitches taken without swinging
`whiffs`	Swing-and-miss count
`swings`	Total swings

Plate discipline captures a dimension of hitting that is largely independent of raw power. A batter with high discipline may not hit many home runs but consistently reaches base, stresses pitchers, and extends at-bats.

Why These Three Targets?

Each target variable measures a fundamentally different dimension of batter performance. Together, they provide a multi-angle picture of what it means to be an effective MLB hitter:

Target	Dimension	Question Answered
`Rendimiento_labels`	Volume Production	How much offensive value does this batter produce overall?
`elite_hitter`	Peak Performance	Does this batter belong among the very best in the league?
`plate_discipline`	Process Quality	How well does this batter make decisions at the plate?

A batter can score differently across all three dimensions. A high-contact, low-power hitter might rank Medio on overall performance, 0 on elite status, but Alta on plate discipline. A slugger might be Alto and elite, but only Media on discipline if they chase breaking balls. Modeling these targets separately lets the project uncover which Statcast features drive each dimension.

Label Distribution Summary

# Overall Performance
print(df["Rendimiento_labels"].value_counts())
# Bajo     ~286
# Medio    ~286
# Alto     ~285

# Elite Status
print(df["elite_hitter"].value_counts())
# 0    684
# 1    173

# Plate Discipline
print(df["clase_disciplina_home"].value_counts())
# Media    (majority)
# Baja
# Alta

Rendimiento_labels is intentionally balanced (equal quantile splits), while elite_hitter is intentionally imbalanced (top 20% threshold). clase_disciplina_home uses fixed score bins, so its distribution reflects the natural spread of discipline scores across the batter population.

Overview

Data

Analysis & Models

Results

Classification Targets: Three Ways to Measure Batter Performance

Rendimiento_labels — Overall Performance

Why wOBA?

Label Construction

Classes

elite_hitter — Elite Status

Why Binary?

Label Construction

Class Distribution

plate_discipline — Plate Decision Quality

Composite Score

Classes

Key Input Metrics

Why These Three Targets?

Label Distribution Summary

Build docs developers (and LLMs) love

Overview

Data

Analysis & Models

Results

Documentation Index

​Rendimiento_labels — Overall Performance

​Why wOBA?

​Label Construction

​Classes

​elite_hitter — Elite Status

​Why Binary?

​Label Construction

​Class Distribution

​plate_discipline — Plate Decision Quality

​Composite Score

​Classes

​Key Input Metrics

​Why These Three Targets?

​Label Distribution Summary

Build docs developers (and LLMs) love

Rendimiento_labels — Overall Performance

Why wOBA?

Label Construction

Classes

elite_hitter — Elite Status

Why Binary?

Label Construction

Class Distribution

plate_discipline — Plate Decision Quality

Composite Score

Classes

Key Input Metrics

Why These Three Targets?

Label Distribution Summary