Investigation on PCA and LDA Feature Extraction Algorithms with KNN Classifier Targetting to Improve Diagnostic Accuracy into Detection of Breast Cancer Malignancy

This work aims to compare two different Feature Extraction Algorithms (FEAs) viz. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA), using a K-Nearest Neighbor (KNN) classifier on the breast cancer Wisconsin (diagnostic) dataset, taken from the UC Irvine machine learning repository. The dataset was divided into a training and test dataset, by selecting 50, 60, 70 and 80% of the whole samples present into the dataset, respectively as the training dataset. The execution was deployed in ‘Google Colab Notebook’ and ‘Anaconda Navigator’, supporting Python 3.6 programming language. For both models, the most accurate values were found for 80% of the training dataset level. PCA provided the best accuracy as 97.37%. A total of 9 quality performance metric parameters such as sensitivity, specificity, positive and negative likelihood ratios, positive and negative predictive values, diagnostic odds ratio, Youden's index and F-1 score, were also obtained for the 80% training dataset level.