An Empirical Evaluation of the t-SNE Algorithm for Data Visualization in Structural Engineering

A fundamental task in machine learning involves visualizing high-dimensional\ndata sets that arise in high-impact application domains. When considering the\ncontext of large imbalanced data, this problem becomes much more challenging.\nIn this paper, the t-Distributed Stochastic Neighbor Embedding (t-SNE)\nalgorithm is used to reduce the dimensions of an earthquake engineering related\ndata set for visualization purposes. Since imbalanced data sets greatly affect\nthe accuracy of classifiers, we employ Synthetic Minority Oversampling\nTechnique (SMOTE) to tackle the imbalanced nature of such data set. We present\nthe result obtained from t-SNE and SMOTE and compare it to the basic approaches\nwith various aspects. Considering four options and six classification\nalgorithms, we show that using t-SNE on the imbalanced data and SMOTE on the\ntraining data set, neural network classifiers have promising results without\nsacrificing accuracy. Hence, we can transform the studied scientific data into\na two-dimensional (2D) space, enabling the visualization of the classifier and\nthe resulting decision surface using a 2D plot.\n