Criminal detection has undergone significant evolution since its inception. It began with basic fingerprinting and manual identification techniques. These days, we rely on advanced biometric systems. In the past, law enforcement used body measurements and later adopted fingerprints to identify criminals. However, these traditional methods had their limitations. They took a lot of time and couldn’t handle big databases. Later on, digital photos and computer vision changed everything. They made it possible to use facial recognition systems that could search huge databases. This study shows a smart criminal detection system. It uses advanced face recognition and machine learning. The approach used, also known as an Enhanced Ensemble method, combines eight different models: k-Nearest Neighbours (k-NN), Support Vector Machine (SVM), Random Forest, XGBoost, Vision Transformer, EfficientNet, ResNet, and DenseNet. Using multiple models lets us compare different ways of solving the problem. It’s a mix of old-school machine learning with cutting-edge deep learning networks in the algorithm. The ensemble method combines the best aspects of each model, which helps address the weak spots by having the models collaborate to make informed decisions. Our enhanced ensemble model works well. It got accuracy rates of 96.7% on the first dataset (D1), 94.1% on the second dataset (D2) and 94.8% on the third dataset (D3). These results show that our combined approach works great for identifying criminals in real-world situations.