Fraud in insurance companies has grown to be a serious problem for the auto insurance sector in recent years, costing honest policyholders a lot of money in lost revenue and raising premiums. The paper offers a novel solution to this problem that uses algorithms in machine learning (ML) and natural language processing (NLP) methods to detect insurance fraud in auto sector. This project also uses AES Encryption technique for storing all the customer details. To find fraudulent patterns in the insurance documents, a machine learning model is constructed applying the proper techniques for XGBoost, logistic regression, Naive Bayes, Random forest, and SVM. To increase accuracy and efficacy, these models are trained using historical data from both valid and fraudulent insurance claims. The models are trained using historical data encompassing both legitimate and fraudulent claims to enhance accuracy and effectiveness. The objective is to compare and determine the algorithm that most accurately differentiates between fraudulent and legitimate claims. Model performance is evaluated using standard metrics such as accuracy, precision, recall, and F1-score. Among all tested algorithms, XGBoost achieved the highest accuracy of 89.73%, demonstrating superior performance in fraud detection.