Generative models are unsupervised learning methods in machine learning that involves automatically discovering and learning the rules or patterns in input data, so that this model can be used to generate new samples. Generative models are an important set of deep neural networks aiming to learn the density function and generate a sample similar to the original data. Variational autoencoder is one of the approaches of Generative models. Variational autoencoder is basically a autoencoder whose encoding distribution is regulated during the training to be ensured that its latent space possesses the features in order for data be generated by the decoder. This regular space allows the generated data to be similar to the original data as possible. In this study, the data generated by the variational autoencoder are clustered by seven algorithms: K-Means, k-Modes, FuzzyCmeans, Agglomerative hierarchical clustering, Mean shift, Spectral clustering and DBSCAN. Additionally, the effect of variational autoencoder on each algorithm was investigated. The data set used in the study is that of students of Isfahan University of Technology. These data are those of the physical health record of the students of Isfahan University of Technology which are answered by the new incoming students as questionnaires every year. The data of this study include the students’ physical, mental, behavioral and medical information and family medical records. Based on lifestyle features, we clustered the students using the K-Means algorithm. We applied the Davis Bouldin index and the Dunn index to eva‎luate clustering algorithms. According to the mentioned indexes, the algorithms K-Means, Fuzzy-Cmeans, Agglomerative hierarchical clustering and spectral hierarchical clustering clustered the data generated by the variational autoencoder more appropriately comparing with the students’ initial data set. Finally; using the algorithm K-Means, three clusters were obtained analyzed based on the features of students’ lifestyles. Obviously, analyzing student data can help university officials plan better.

الهام محمودزاده

محمد رضا احمد زاده، عليرضا بصيري، سمانه حسيني