Automatic Classification of Banking Branch Requests and Errors with Natural Language Processing and Machine Learning

: In today’s world, the service sector is undergoing rapid technological development. Keeping pace with this transformation is not only a necessity for institutions but also essential for gaining a competitive advantage. Technological advancements have made it crucial to respond to customer demands quickly and accurately.This study emphasizes the importance of efficiently classifying customer demands and software errors in the branches of a bank operating in the finance sector. Real-world data was divided into three categories: Desktop Support, Software Support, and Field Support, with a total of 4,500 samples equally distributed among the categories. Eighty percent of the data was used for training and 20% for testing machine learning algorithms such as Bidirectional Encoder Representations from Transformers (BERT), Naive Bayes, Random Forest, and Artificial Neural Networks (ANN). The models were trained separately using CountVectorizer and Term Frequency–Inverse Document Frequency (TF-IDF) metrics. The dataset was also analyzed using two sample sizes: 3,000 and 4,500. The best results were obtained with BERT and ANN models using 4,500 samples and the TF-IDF metric, achieving accuracy rates above 92%. The positive effects of increased data size and the TF-IDF metric were evident. Additionally, ANN-based models proved more effective for this type of classification problem.