Skip to main content

​
Overview

The feature engineering module creates derived features from cleaned hospital data, including age ranges, adult indicators, and BMI risk categories.

​
Functions

​
build_features

Builds derived features from cleaned hospital data for use in predictive modeling. 
def build_features(df: pd.DataFrame) -> pd.DataFrame
​
df
pd.DataFrame
required
Cleaned hospital DataFrame. Typically the output from clean_hospital_data().
​
return
pd.DataFrame
DataFrame with all original columns plus three new engineered features:
  • age_range: Categorical age range (“0-15”, “15-35”, “35-55”, “55-70”, “70-80”)
  • is_adult: Binary indicator (1 if age >= 18, else 0)
  • bmi_risk: BMI risk category (0=underweight, 1=normal, 2=overweight, 3=obese)
Example: 
import pandas as pd
from feature_engineering.features import build_features

# Assume clean_df comes from clean_hospital_data()
feature_df = build_features(clean_df)

print(f"Original columns: {len(clean_df.columns)}")
print(f"After feature engineering: {len(feature_df.columns)}")
print(f"New features: age_range, is_adult, bmi_risk")
print(f"\nAge range distribution:\n{feature_df['age_range'].value_counts()}")
print(f"\nAdult ratio: {feature_df['is_adult'].mean():.2%}")
print(f"\nBMI risk distribution:\n{feature_df['bmi_risk'].value_counts()}")

​
Engineered Features

​
age_range

Categorizes patient age into clinical age ranges. Type: Categorical (ordinal) Categories:
RangeLabelDescription
[0, 15)“0-15”Pediatric
[15, 35)“15-35”Young adult
[35, 55)“35-55”Middle age
[55, 70)“55-70”Senior
[70, 80]“70-80”Elderly
Implementation: 
feat["age_range"] = pd.cut(
    feat["age"], 
    bins=[0, 15, 35, 55, 70, 80], 
    labels=["0-15", "15-35", "35-55", "55-70", "70-80"], 
    right=False
)
Example Output: 
# Patient aged 42
age_range = "35-55"

# Patient aged 68
age_range = "55-70"

# Patient aged 12
age_range = "0-15"

​
is_adult

Binary indicator for adult patients (age >= 18). Type: Integer (0 or 1) Values:
  • 1: Patient is an adult (age >= 18)
  • 0: Patient is a minor (age < 18)
Implementation: 
feat["is_adult"] = (feat["age"] >= 18).astype(int)
Example Output: 
# Patient aged 25
is_adult = 1

# Patient aged 16
is_adult = 0

# Patient aged 18 (exactly)
is_adult = 1
Use Cases:
  • Filtering adult-only analyses
  • Stratifying models by age group
  • Computing age-specific metrics
  • Regulatory compliance (pediatric vs. adult protocols)

​
bmi_risk

Body Mass Index risk category based on clinical thresholds. Type: Float (0.0, 1.0, 2.0, or 3.0) Categories:
BMI RangeRisk LevelCodeDescription
< 18.5Underweight0.0Below healthy weight
[18.5, 25)Normal1.0Healthy weight range
[25, 30)Overweight2.0Above healthy weight
>= 30Obese3.0Clinically obese
Implementation: 
feat["bmi_risk"] = pd.cut(
    feat["bmi"], 
    bins=[-1, 18.5, 25, 30, 100], 
    labels=[0, 1, 2, 3]
).astype(float)
feat["bmi_risk"] = feat["bmi_risk"].fillna(0)
Example Output: 
# Patient with BMI 22.5
bmi_risk = 1.0  # Normal

# Patient with BMI 28.3
bmi_risk = 2.0  # Overweight

# Patient with BMI 17.0
bmi_risk = 0.0  # Underweight

# Patient with BMI 32.5
bmi_risk = 3.0  # Obese
Missing Value Handling: Any missing or invalid BMI values are filled with 0.0 (underweight category).

​
Constants

​
AGE_BINS

Bin edges for age range categorization. 
AGE_BINS = [0, 15, 35, 55, 70, 80]
Defines 5 age ranges with boundaries at 0, 15, 35, 55, 70, and 80 years.

​
AGE_LABELS

Labels for age range categories. 
AGE_LABELS = ["0-15", "15-35", "35-55", "55-70", "70-80"]
Human-readable labels corresponding to AGE_BINS.

​
Complete Pipeline Example

from pathlib import Path
from ingestion.loader import load_hospital_data, merge_hospital_data
from preprocessing.cleaning import clean_hospital_data
from feature_engineering.features import build_features

# Complete data preparation pipeline
data_dir = Path("data/")

# 1. Load data
datasets = load_hospital_data(data_dir)
print(f"Loaded {len(datasets)} datasets")

# 2. Merge data
merged = merge_hospital_data(datasets)
print(f"Merged to {len(merged)} records")

# 3. Clean data
clean = clean_hospital_data(merged)
print(f"Cleaned {len(clean)} records")

# 4. Build features
features = build_features(clean)
print(f"Built {len(features.columns)} total columns")
print(f"New features: age_range, is_adult, bmi_risk")

# 5. Analyze feature distributions
print("\nFeature Statistics:")
print(f"Adult patients: {features['is_adult'].sum()} ({features['is_adult'].mean():.1%})")
print(f"\nAge range distribution:")
print(features['age_range'].value_counts().sort_index())
print(f"\nBMI risk distribution:")
print(features['bmi_risk'].value_counts().sort_index())

# 6. Use features for modeling
feature_cols = ['age', 'bmi', 'age_range', 'is_adult', 'bmi_risk']
X = features[feature_cols]
print(f"\nFeature matrix shape: {X.shape}")

​
Feature Engineering Best Practices

  1. Always apply to cleaned data: Run clean_hospital_data() before build_features()
  2. Preserve original data: Features are added, not replaced
  3. Handle edge cases: Missing BMI values default to 0.0 (underweight)
  4. Use consistent bins: Age and BMI bins align with clinical standards
  5. Document thresholds: All cutoff values are defined as module constants
  6. Type consistency: BMI risk is float to allow NaN handling before filling

Build docs developers (and LLMs) love

Get started for freeTalk to us