Abstract

This paper presents a novel approach to revolutionize customer complaint handling through a machine learning-based ensemble classification system and an automated escalation mechanism. The research involves the development of an ensemble model, integrating random forest, XGBoost, and Support Vector Machine (SVM), selected from a thorough evaluation of 10 machine learning models. The objective is to streamline the classification of customer emails into predefined categories—specifically, "replacements and refunds issues", "account related issues", and "others". The methodology involves the extraction and manual labelling of approximately 2000 emails that were obtained after web scraping. The email bot incorporates an algorithm to determine urgency levels based on factors such as product cost, complaint history, product importance, customer sentiment, and keyword analysis. Integrated with the Gmail API, the bot efficiently tags and escalates incoming emails to the relevant department. Performance metrics, including total resolution time and write cycles, gauge the success of the implemented solution. Results indicate a significant reduction in the time and effort required for complaint resolution, thereby enhancing overall customer experience. This research emphasizes the advantages of leveraging machine learning to optimize customer support processes, paving the way for future advancements in this field.