Automatic Complaints Classification in E-Commerce: A Case Study Using CRISP-DM

The growth of e-commerce has been remarkable in recent years, driven by increasing consumer demand for attention and quick responses. Given the large volume of transactions and complaints accompanying this increase, automating the classification of these complaints can help quickly route them to the appropriate departments. This paper presents a computational approach using the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology to automate the complaints screening process. We categorized 600 real complaints from three e-commerce platforms in Brazil. The learning model was trained progressively, using an initial set of 25 complaints in each category. The classification model obtained an accuracy of 85% and an average of over 80% across all relevant metrics, including precision, recall, and F1-Score. The results confirmed the effectiveness of the developed model for automated complaint classification in e-commerce, providing a computational strategy that improves the customer service process and allows for quicker problem resolution.