Skip to main content
The feature engineering module creates new features from cleaned hospital data to improve model performance and enable better analysis. It transforms continuous variables into categorical ranges and computes derived indicators.

​
Building Features

The build_features() function creates three new features from the cleaned dataset.

​
Function Signature

def build_features(df: pd.DataFrame) -> pd.DataFrame
Parameters:
  • df (pd.DataFrame): Cleaned hospital data from the preprocessing pipeline
Returns:
  • DataFrame with original columns plus new engineered features

​
Engineered Features

​
1. Age Range

Categorizes patients into age groups using pd.cut(): 
AGE_BINS = [0, 15, 35, 55, 70, 80]
AGE_LABELS = ["0-15", "15-35", "35-55", "55-70", "70-80"]

feat["age_range"] = pd.cut(
    feat["age"], 
    bins=AGE_BINS, 
    labels=AGE_LABELS, 
    right=False
)
Age groups:
  • 0-15 - Children and adolescents
  • 15-35 - Young adults
  • 35-55 - Middle-aged adults
  • 55-70 - Older adults
  • 70-80 - Senior adults
Parameters:
  • right=False - Left-inclusive bins (e.g., 15 ≤ age < 35)
Example: 
age = 42
age_range = "35-55"  # 35 <= 42 < 55

​
2. Adult Indicator

Binary flag indicating whether the patient is an adult (≥18 years): 
feat["is_adult"] = (feat["age"] >= 18).astype(int)
Values:
  • 1 - Adult (age ≥ 18)
  • 0 - Minor (age < 18)
Example: 
age = 16 → is_adult = 0
age = 18 → is_adult = 1
age = 45 → is_adult = 1

​
3. BMI Risk Category

Categorizes Body Mass Index into risk levels based on standard health guidelines: 
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)
Risk levels:
  • 0 - Underweight (BMI < 18.5) or missing
  • 1 - Normal weight (18.5 ≤ BMI < 25)
  • 2 - Overweight (25 ≤ BMI < 30)
  • 3 - Obese (BMI ≥ 30)
BMI ranges:
BMI RangeRisk LevelCategory
< 18.50Underweight
18.5 - 24.91Normal
25 - 29.92Overweight
≥ 303Obese
Example: 
bmi = 22.3 → bmi_risk = 1.0  # Normal
bmi = 27.8 → bmi_risk = 2.0  # Overweight
bmi = 31.5 → bmi_risk = 3.0  # Obese
bmi = 0.0  → bmi_risk = 0.0  # Missing (filled with 0)

​
Usage Example

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

# Complete pipeline
datasets = load_hospital_data(data_dir)
merged = merge_hospital_data(datasets)
clean = clean_hospital_data(merged)

print(f"Before feature engineering: {clean.shape[1]} columns")

# Build features
feat = build_features(clean)

print(f"After feature engineering: {feat.shape[1]} columns")
print(f"New features: {['age_range', 'is_adult', 'bmi_risk']}")
From cli.py:50-53: 
datasets = load_hospital_data(CONFIG.data_dir)
merged = merge_hospital_data(datasets)
clean = clean_hospital_data(merged)
feat = build_features(clean)

​
Feature Output Example

​
Input (Cleaned Data)

   age  gender    bmi  height  weight
0  45   m        28.3  175     87
1  16   f        19.2  162     50
2  67   m        31.2  170     90
3  25   f        22.1  168     62

​
Output (With Engineered Features)

   age  gender    bmi  height  weight  age_range  is_adult  bmi_risk
0  45   m        28.3  175     87      35-55      1         2.0
1  16   f        19.2  162     50      15-35      0         1.0
2  67   m        31.2  170     90      55-70      1         3.0
3  25   f        22.1  168     62      15-35      1         1.0

​
Pipeline Integration

Feature engineering connects cleaning to modeling: 
from ingestion.loader import load_hospital_data, merge_hospital_data
from preprocessing.cleaning import clean_hospital_data
from feature_engineering.features import build_features
from modeling.predictive import train_predictive_models

# Full pipeline
datasets = load_hospital_data(CONFIG.data_dir)
merged = merge_hospital_data(datasets)
clean = clean_hospital_data(merged)
feat = build_features(clean)  # Feature engineering step

# Use features for modeling
artifacts = train_predictive_models(
    feat, 
    CONFIG.feature_columns, 
    CONFIG.target_risk, 
    CONFIG.target_outcome
)

​
Feature Analysis

Analyze the distribution of engineered features: 
feat = build_features(clean)

# Age range distribution
print("Age Range Distribution:")
print(feat["age_range"].value_counts().sort_index())

# Adult vs minor counts
print("\nAdult Status:")
print(f"Adults: {feat['is_adult'].sum()}")
print(f"Minors: {(feat['is_adult'] == 0).sum()}")

# BMI risk distribution
print("\nBMI Risk Distribution:")
print(feat["bmi_risk"].value_counts().sort_index())
Example output: 
Age Range Distribution:
0-15      120
15-35     450
35-55     380
55-70     210
70-80      40

Adult Status:
Adults: 1080
Minors: 120

BMI Risk Distribution:
0.0    150
1.0    520
2.0    380
3.0    150

​
Constants Reference

The module defines these configuration constants: 
AGE_BINS = [0, 15, 35, 55, 70, 80]
AGE_LABELS = ["0-15", "15-35", "35-55", "55-70", "70-80"]
Import them for consistency in analysis: 
from feature_engineering.features import AGE_BINS, AGE_LABELS

print(f"Age bins: {AGE_BINS}")
print(f"Age labels: {AGE_LABELS}")

​
Feature Selection for Modeling

The engineered features can be used alongside original columns: 
# Original features
original_features = ['age', 'height', 'weight', 'bmi', 'children', 'months']

# Engineered features
engineered_features = ['age_range', 'is_adult', 'bmi_risk']

# Combined feature set
all_features = original_features + engineered_features

# Select features for modeling
X = feat[all_features]
y = feat['diagnosis']
From cli.py:55-59: 
artifacts = train_predictive_models(
    feat, 
    CONFIG.feature_columns,  # Configured feature list
    CONFIG.target_risk, 
    CONFIG.target_outcome
)

Build docs developers (and LLMs) love

Get started for freeTalk to us