<p>Fraud in the insurance sector represents a significant challenge for companies and society, leading to substantial economic losses for insurers and impacting pricing policies, ultimately affecting the users of these insurance services. This study highlights the urgent need to develop effective methods for fraud detection, especially in contexts where data is imbalanced and fraudulent cases are much less frequent than non-fraudulent ones. The primary objective of this study is to design and validate a fraud detection model for the insurance sector, with a specific focus on automobile insurance claims. The methodology used to implement the Random Forest Quantile Classifier is detailed, emphasizing its ability to optimize classification thresholds, achieving the best possible specificity and sensitivity. This model also allows for adjusting these thresholds to meet the specific needs of the insurance company. A case study was conducted using actual data provided by an insurance company to evaluate the performance of the proposed model. The performance of the Random Forest Quantile Classifier was compared with other machine learning methods. The results indicated that the proposed model outperformed the others in fraud detection, achieving an optimal balance between true positive and true negative rates. Additionally, the model includes the ability to interpret the effects of variables on predictions, preventing it from functioning as a black box. This research contributes to advancing and improving fraud detection techniques, providing valuable tools for professionals and researchers dedicated to financial data analysis.</p>