New Artificial intelligence approaches for brand switching decisions

Abstract

The problem of customer complaints occurs in almost every business and solutions are offered to reduce these complaints. When companies do not pay necessary attention to the complaints, they suffer revenue losses due to the loss of customers, the brand and the company's image. In this study, based on a website data holding customer complaints, the customers’ decisions about brand switching are predicted. Different machine learning (ML) and two newly proposed deep learning (DL) techniques are used and compared in terms of their classification performance. Prediction results of the various ML and DL algorithms are analyzed and the test metrics values obtained from k-fold cross-validation and the train/test split methods are presented. The ML and DL techniques were used to classify the original dataset and the dataset obtained after data augmentation. Linear Discriminant Analysis (LDA) and proposed ResNet2 achieved the best performance with 88.30% and 90.83% test accuracy for the original and the augmented dataset, respectively. Different non-parametric statistical methods were used to test for mean differences between ordinal variables. The results show that the best and/or comparable results are achieved when proposed DL methods are employed by using k-fold-cross-validation technique on augmented data sets.