A Comparative Dimensionality Reduction Study in Diabetes Patients using LDA and PCA

In electronic life, we deal with huge data daily in different sectors like the medical sector, social media, the economic sector, etc. Data is streaming currently every day Due to the transformation of all transactions at various levels to electronic dealing, this led to the emergence of the so-called curse of dimensions, which is mostly appearing in big data that have too many of attributes (Features), To avoid this problem and in order to take advantage of the huge amount of available data and contribute to decision-making at various levels, machine learning techniques have been used. In this article, we used Pima Indians Diabetes Database from www.kaggel.com source of this data from the “National Institute of Diabetes and Digestive and Kidney Diseases”, The porous of this study is a comparison between (Linear Discriminant Analysis (LDA) and Principal Component Analysis (PCA)) performance after classification process, and detecting each classifier efficiency With changing the number of features, four classifiers used in the experiment (Naive Bayes, Random Forest, SVM, and KNN) to predict the patient case if he has diabetes or not, We removed some features by dimensionality reduction techniques, (LDA) and (PCA), the accuracy of prediction before and after reduction has been evaluated, results illustrate that Support vector machine(SVM)and Naïve Bayes achieved the best accuracy by (82%). also, we touched on to the performance of (PCA and LDA). The results showed that LDA performance was better than PCA in prediction.