The study will develop a sophisticated behavioral biometric system for secure user verification through the application of AI driven gait recognition. The project will explore the generation of synthetic data to improve the performance and resilience of a model, as opposed to simply augmenting existing datasets. An evaluation of five state-of-the-art deep learning models developed in both TensorFlow/Keras and PyTorch has been conducted to examine their performance. The five models are: ResNet-50 Attention (accuracy of 92.34%), Vision Transformer (ViT) (accuracy of 94.56%), Bidirectional LSTM CNN (accuracy of 91.23%), EfficientNet Temporal Fusion (accuracy of 93.78%), and an ensemble of multiple models. All five models have demonstrated high levels of accuracy, with the highest accuracy achieved by the ensemble model at 95.67%. Additionally, the study has shown that synthetic data is highly effective and the average increase in performance achieved when utilizing synthetic data was 15.2%, compared to models trained solely on real data. Furthermore, this approach improved not only the accuracy of the models, but also the generalizability of the models and prevented the issue of overfitting. Finally, the study has shown that there exists a slight advantage to the average accuracy (average of 94.67%) of PyTorch models compared to TensorFlow/Keras models (average of 91.79%). This difference is statistically significant (p < 0.05), and thus the study presents a comprehensive solution for gait recognition and demonstrates the applicability of these models for secure and efficient user authentication in real world environments. Ultimately, the study confirms that the combination of ensemble methodologies and synthetic data generation is the current best practice for gait-based biometric authentication.