Data balancing helps prevent biased results by ensuring the training data adequately represents all classes.

• A. Data leakage
• B. Bias due to class imbalance
• C. Overfitting due to excessive features
• D. Underfitting due to insufficient data

Answer: (B)

Focusing on bias that comes from uneven representation among classes explains why data balancing is used. When one class dominates the training data, the model may learn to favor that class and perform poorly on minority classes, giving a misleading sense of accuracy. Balancing the data—through methods like resampling, creating synthetic minority examples, or using class-weighted learning—helps the model see enough examples of every class so its predictions are fairer and more accurate across all categories.

Data leakage is about information from the test data slipping into training, which inflates performance but isn’t about class representation. Overfitting due to too many features arises when the model memorizes training data rather than generalizing, not specifically from imbalanced classes. Underfitting due to too little data happens when there isn’t enough information for the model to learn patterns, also separate from class balance concerns.