Insurance fraud poses a significant challenge to the industry, resulting in substantial financial losses each year. This paper investigates the application of various ML models to detect fraudulent insurance claims, including Logistic Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), and Gradient Boosting (GB). Additionally, ensemble techniques such as stacking and majority voting are explored to improve detection accuracy. The results demonstrate that the stacked model outperforms individual models, achieving an accuracy of 0.90 and an AUC-ROC of 0.93. While RF and GB performed well individually, the stacked model provided a more balanced solution, reducing both false positives and false negatives. The findings highlight the importance of model ensembling in improving fraud detection and suggest future work in resampling techniques to address imbalanced data. These results offer practical implications for the insurance industry, where accurate fraud detection is essential to reduce financial risks.