The Machine Learning “Advent Calendar” Day 3: GNB, LDA and QDA in Excel

Three classification models are explained and compared: Gaussian Naive Bayes (GNB), Linear Discriminant Analysis (LDA), and Quadratic Discriminant Analysis (QDA). These models build upon simpler concepts like nearest centroids by incorporating statistical properties such as variance and covariance to define class shapes. LDA uses a shared covariance matrix across classes, resulting in linear decision boundaries, whereas QDA uses a separate covariance matrix for each class to create quadratic boundaries. The implementation and visualization of these algorithms, including the calculation of Mahalanobis distance and class probabilities, are demonstrated step-by-step in Excel.