Discriminant Analysis

This chapter explains the concept of procedure to undertake discriminant analysis. Explaining its difference and similarity with logistic regression, we begin with explaining the process for preparing the data for undertaking discriminant analysis, using the mtcars dataset. Before undertaking discriminant analysis, we first test the assumptions of homogeneity of variance, and multicollinearity for the predictors. In discriminant analysis, the outcome is required to be categorical (dichotomous or polytomous). We then explain the procedure to undertake linear discriminant analysis (LDA) and generate a plot based on the classification of cars, categorised as High, Medium and Low horsepower cars, based on two linear discriminants generated from six predictors. However, the plot shows that the classification is more likely to be quadratic rather than linear. We, therefore, explain the procedure to undertake quadratic discriminant analysis (QDA). Finally, we discuss the concept of regularised discriminant analysis, a combination of linear and quadratic approaches. Despite more robust compared to individual LDA and QDA, RDA also does not generate values of two discriminants, though predict() can be used for classification purposes.