How to Handle Imbalanced Data

When one class dominates your dataset (e.g., 99% Non-Fraud, 1% Fraud), standard ML algorithms bias entirely toward the majority class.

Why Imbalance is Dangerous

If a model blindly predicts "Non-Fraud" for every single transaction, it achieves 99% Accuracy. However, detecting fraud is the entire business objective. Standard algorithms minimise overall error, sacrificing the minority class entirely to achieve high global accuracy.

Method 1: Algorithmic Class Weights

The simplest solution is adjusting class_weight. Most scikit-learn classifiers accept "balanced", which automatically increases the penalty for misclassifying minority observations.

import seaborn as sns
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report

df = sns.load_dataset("diamonds").sample(1000, random_state=42)
X = df[["carat", "depth", "table"]]
y = (df["cut"] == "Premium").astype(int)

X_tr, X_te, y_tr, y_te = train_test_split(X, y, random_state=42)

model = LogisticRegression(class_weight="balanced", max_iter=1000)
model.fit(X_tr, y_tr)
print(classification_report(y_te, model.predict(X_te)))

Method 2: Synthetic Minority Oversampling (SMOTE)

Instead of reweighting the algorithm, you synthetically generate new minority-class data points by interpolating between existing minority observations.

You need to install the imbalanced-learn library: pip install imbalanced-learn.

from imblearn.over_sampling import SMOTE
from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)

smote = SMOTE(random_state=42)
X_train_res, y_train_res = smote.fit_resample(X_train, y_train)

print(f"Original training shape: {X_train.shape}")
print(f"Resampled training shape: {X_train_res.shape}")

Critical Rule

Never apply SMOTE before train_test_split. If you synthesise data before splitting, synthetic copies of minority observations will leak into the test set, giving you artificially inflated scores that do not reflect real-world performance.

Method 3: Threshold Tuning

By default, classifiers use a 0.5 probability threshold. For imbalanced problems, lowering the threshold (e.g., to 0.3) increases Recall at the cost of Precision.

import numpy as np

probs = model.predict_proba(X_te)[:, 1]
custom_preds = (probs >= 0.3).astype(int)
print(classification_report(y_te, custom_preds))

KSB Mapping

KSB	Description	How This Addresses It
K4.1	Statistical models and methods	Understanding the statistical basis of regression and classification
K4.2	ML and AI techniques	Implementing and comparing supervised learning algorithms
K4.4	Resource constraints and trade-offs	Model complexity vs interpretability; computational cost
S1	Scientific methods and hypothesis testing	Formulating hypotheses and testing with rigorous validation
S4	Building models and validating	Cross-validation, train/test evaluation, performance metrics
B5	Impartial, hypothesis-driven approach	Honest evaluation of model performance and limitations