Outlier Detection in High-Dimensional Data using t-Distributed Stochastic Neighbor Embedding

In this study, the focus is to use the t-distributed Stochastic Neighbor Embedding (t-SNE) technique for identifying outliers from high dimensional data sets. To this end, the study proposes to use t-SNE with other clustering algorithms with the view of improving upon the performance of outlier detection. The percentage prognostications were based on the results obtained by applying the proposed methodology on three sets: CIFAR-10, KDD Cup 1999, and HAR; the last one yielded an accuracy of 89%. 5%, 92. 3%, and 93. With K-Means clustering, it was possible to make predictions of the number of patients for the next year with a 95% accuracy while the patients’ age was clustered into two groups with an 8% error and gender into two groups with an 8% error. The findings show that widely acclaimed t-SNE can be used to identify clusters and anomalies in large datasets. In future works additional, efforts will be made to determine optimal values of t-SNE parameters and, also, to investigate the possibilities of further combining of t-SNE with various machine learning models.