This repository contains research on dimensionality reduciton algorithms for the classification task. Chosen methods are: PCA, KPCA, SPCA, LDA and own implementation of L1PCA*. They were compared on 5 different datasets and 4 popular classifiers (kNN, GNB, SVM and CART). The goal is to find the best algorithm for handling dimensionality reduction in classification tasks.
The experiment was carried out on both synthetic and real data. The experiment protocol as well as results for real data can be found in src/pca_classification_real.ipynb file. Based on statystical analysis (t-student test) LDA based algorithms were significantly better than algorithms based on PCA on the given datasets. It is worth nothing that the conducted experiment is a small-scale trial. A good practice would be to repeat it using a larger range of datasets.