This paper presents a CNN system for detecting healthcare fraud using multivariate time series claims data. The model automatically extracts spatial features from raw time series inputs, and is trained on a large dataset containing claims data, including diagnosis codes, procedures, providers, costs, and fraud labels. Hyper parameter optimization is performed to enhance fraud detection performance. The convolutional neural network system achieves an F1 score of 0.82, outperforming traditional methods such as gradient boosting machines and random forests. To improve transparency, layer-wise relevance propagation is used to identify key features influencing each prediction, thereby highlighting anomalous provider activities suggestive of fraud. The proposed system combines advanced fraud detection with feature relevance explanations, promoting trust and auditability, and demonstrating the value of explainable deep learning in responsible healthcare fraud detection.