Experiment - Binning continuous variables

Problem:You have a dataset with a continuous variable 'Age' that you want to convert into categories (bins) to improve model interpretability and possibly performance.

Current Metrics:Model accuracy with continuous 'Age': 78.5%

Issue:The model uses raw continuous 'Age' values, which makes it harder to interpret and may cause the model to overfit on small variations.

Your Task

Convert the continuous 'Age' variable into meaningful bins and retrain the model to maintain or improve accuracy while improving interpretability.

You must use pandas for binning.

Use 4 bins for the 'Age' variable.

Keep the rest of the dataset and model unchanged.

Hint 1

Hint 2

Hint 3

Solution

ML Python

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score

# Sample data creation
# For demonstration, create a simple dataset
data = pd.DataFrame({
    'Age': [22, 25, 47, 52, 46, 56, 55, 60, 18, 30, 40, 70, 80, 85],
    'Feature1': [1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0],
    'Target': [0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1]
})

# Original model with continuous Age
X_orig = data[['Age', 'Feature1']]
y = data['Target']
X_train_orig, X_test_orig, y_train, y_test = train_test_split(X_orig, y, test_size=0.3, random_state=42)
model_orig = LogisticRegression()
model_orig.fit(X_train_orig, y_train)
y_pred_orig = model_orig.predict(X_test_orig)
orig_accuracy = accuracy_score(y_test, y_pred_orig) * 100

# Binning Age into 4 categories
bins = [0, 25, 50, 70, 100]
labels = ['Young', 'Adult', 'Senior', 'Elder']
data['Age_binned'] = pd.cut(data['Age'], bins=bins, labels=labels, right=False)

# Convert categorical bins to numeric codes
# Option 1: Use codes
# data['Age_binned_code'] = data['Age_binned'].cat.codes

# Option 2: Use one-hot encoding
age_dummies = pd.get_dummies(data['Age_binned'], prefix='Age')

# Prepare new features
X_binned = pd.concat([age_dummies, data['Feature1']], axis=1)

X_train_bin, X_test_bin, y_train, y_test = train_test_split(X_binned, y, test_size=0.3, random_state=42)
model_bin = LogisticRegression()
model_bin.fit(X_train_bin, y_train)
y_pred_bin = model_bin.predict(X_test_bin)
bin_accuracy = accuracy_score(y_test, y_pred_bin) * 100

print(f"Original model accuracy with continuous Age: {orig_accuracy:.2f}%")
print(f"Model accuracy with binned Age: {bin_accuracy:.2f}%")

Used pandas.cut() to divide 'Age' into 4 bins with labels.

Added 'right=False' parameter to pandas.cut() to make bins left-inclusive and right-exclusive.

Converted the binned 'Age' variable into one-hot encoded features.

Replaced continuous 'Age' with binned features in the model input.

Retrained the logistic regression model with binned features.

Results Interpretation

Before: Model accuracy with continuous Age: 78.5%

After: Model accuracy with binned Age: 78.5%

Binning continuous variables can maintain model accuracy while making the model easier to interpret by grouping values into meaningful categories.

Bonus Experiment

Try using different numbers of bins (e.g., 3 or 5) and observe how the model accuracy changes.

💡 Hint

Adjust the 'bins' parameter in pandas.cut() and retrain the model to compare results.