This chapter presents foundational techniques for improving model generalizationGeneralization in supervised learning by addressing challenges such as overfittingOverfit/overfitting and underfitting. It begins with a formal introduction to datasetsDatadataset, including training, testTest, and validationValidation splits, and their respective roles in model evaluation. The chapter then explores key concepts such as Mean Squared ErrorMean squared error (MSE) (MSE), biasBias, and varianceVariance, which are essential for understanding model behavior. Core strategies for mitigatingOverfit/overfitting overfitting—such as cross-validation and regularizationRegularization using \(\ell _1\) and \(\ell _2\) norms—are introduced, followed by advanced regularization methods commonly employed in deepLearningdeep learning learningDeepdeep learning, including weight decayWeightweight decay, noise injectionNoisenoise injection, early stoppingEarly stopping, dropoutDropout, and batch normalizationBatchbatch normalization. The chapter also discusses ensembleEnsemble methods like bagging, offering theoretical insights into why these approaches improve generalizationGeneralization performance.