Impact of Data Breadth and Depth on Performance of Siamese Neural Network Model: Experiments with Two Behavioral Biometric Datasets

Deep learning models, such as the Siamese Neural Networks (SNN), have shown great potential in capturing the intricate patterns in behavioral data. However, the impact of dataset breadth (i.e., the number of subjects) and depth (i.e., the amount of data per subject) on the performance of these models remain unexplored. To this end, we have conducted extensive experiments using two publicly available large datasets (Aalto and BrainRun), varying both the number of training subjects and the number of samples per subject. Our results show that dataset depth plays a crucial role in capturing more intricate variations specific to individual subjects, thereby positively influencing the performance of the SNN models. On the other hand, increasing the dataset breadth enables the model to effectively capture more inter-subject variability, which proved to be a more significant factor in improving the overall model performance. Specifically, once a certain threshold for the number of training subjects is surpassed, breadth starts to dominate performance and the impact of dataset depth diminishes and disappears. These findings shed light on the importance of dataset breadth and depth in training deep learning models for behavioral biometrics and provide valuable insights for designing more effective authentication systems.