The Use of Autoencoders in Reducing The Dimensions of Survey Results

Machine learning is a suitable approach for the analysis of large data sets, their visualization, pattern recognition, as well as much more complex tasks such as speech recognition and natural language processing. The article discusses one of the machine learning algorithms – autoencoder (AE) and its practical application in the analysis of data from student motivation surveys (AMS – Academic Motivation Scale). The results of the survey data reduction to two and three dimensions are presented graphically. Using the k-fold cross validation procedure, the AE-reduced data decoding results were compared to those obtained by principal component analysis (PCA). The results show that AE can be considered as a reliable tool for reducing multidimensional survey data, as well as to detect anomalies in this type of data. The paper concludes that attention should be paid to recommendations related to the use of machine learning methods, including monitoring the learning process to avoid overfitting error and verifying the datasets used for model optimization.