Applied ML engineering for complaint resolution: A case study in data preparation and model comparison

This study focuses on evaluating company trustworthiness in the context of consumer fraud complaints by analyzing patterns within the Consumer Financial Protection Bureau’s public dataset. The objective is to identify which companies respond most effectively to fraud-related complaints, using this behavior as a proxy for consumer satisfaction. The approach involves associating company responses with product categories to determine which organizations perform better in resolving issues. Key phases include extensive data cleaning, handling missing values, and identifying uninformative attributes. Various machine learning algorithms, such as Naïve Bayes, J48, Random Forest, and k-Nearest Neighbors, were applied to classify complaint outcomes. The findings indicate that company responses are highly imbalanced, with a majority of cases resulting in