Fuzzy Logic and Deep Learning Integration in Likert Type Data

Abstract

Deep learning networks have many modern applications and demonstrate a high-performance level. As the applications of deep learning networks to real-world problems continues to spread, the reason why they are effective remains unknown. However, it is possible to make some judgments by examining the behaviour of the network in experiments. The main aim of this study is to analyse the performance of deep learning techniques in the form of a 5-point Likert-type scale by converting the artificial data sets into a fuzzy form using triangular or trapezium fuzzy numbers. To test the performance of the proposed model, which is the integration of deep learning and fuzzy logic techniques, the satisfaction estimation problem was chosen. Data sets consisting of fuzzy numbers which reach at least three or four times more parameters than normal data sets. Thus, it decreases the possibility of falling into the local optimum trap in optimization studies with big data. In the analysis conducted with deep learning, in accordance with the fuzzification examples in the literature, the defuzzification was carried out with separate results for peak, maximum, and minimum values. In contrast to the literature, the performances of the deep learning model were investigated by suggesting that fuzzy numbers produce a single result series.