Supervised Learning Approach towards Class Separability- Linear Discriminant Analysis

Abstract

Feature extraction can be observed as a step involved in preprocessing phase. It helps to remove redundant inconsistent data from a dataset. After this, job of classifiers becomes smooth and they perform better. In supervised algorithms, the extracted or main features are used in categorisation of data to its class. In this paper, we have done experiments on Linear Discriminant Analysis (LDA) which is a technique of dimensionality reduction used in various areas like machine learning and pattern classification. LDA projects datasets on lower-dimensions having larger class-separation which in turn helps to minimise the computational costs and avoid overfitting. The experiments are conducted on various datasets, their performance is checked using various metrics and a graphical view of class separability is also shown for a better understanding to the reader. In our results analysis we have used the Logistic regression classifier for classification of data points to their accurate classes and we have received an accuracy of 100% with wine dataset and 97% accuracy with bank-note dataset. This shows the remarkable efficacy of this algorithm.