Reducing Complexity: A Comparative Analysis of Dimensionality-Reduction Techniques

Machine Learning (ML) is a modern fast-growing technology. It has extensive applications in disciplines like computer vision, bioinformatics, the medical field, finance, fraud detection, and so on. As we’ve seen, real-world data are used to train machine learning models, which can be used to process the data for purposes like prediction, classification, image processing, etc. The number of attributes in a dataset is known as its dimension. Real-world data sets have a lot of variables, which increases their dimension. It becomes increasingly challenging to envision and work on the training set as the number of variables increases. Sometimes the correlation between most of these factors makes them unnecessary. The benefits of using dimensionality are that when it comes to data compression and gaining less data space, dimensionality reduction is a great instrument. It drastically cuts down on computation time and it makes it simple to remove redundant characteristics from datasets. This project aims to provide a comparison of different dimensionality reduction algorithms. Dimensionality reduction uses a variety of techniques that yield varied outcomes. To compare and analyze the same, we have chosen four algorithms: The linear discriminant analysis algorithm (LDA), the CUR matrix decomposition algorithm, the singular value decomposition algorithm (SVD), and the principal component analysis algorithm (PCA). We are assessing the performance of these algorithms using the classifiers: Decision Tree classifier, Gaussian Naïve Bayes classifier, and the Random-Forest classifier, based on the parameter’s accuracy, recall, precision, and F-measure.