Apparent Personality Traits Detection Based on Correlation-Based Attention and Feature Weighting Methods

Apparent personality traits (APT) is an impression that individual leaves to others. Current work used publicly accessible Video Dataset for Personality Traits detection (VDPT) to propose end-to-end machine learning solution for APT prediction. These predictions are based on a correlation-based attention mechanism for effectively assigning weights to facial features. We explored three research questions in this work. They are: First, how does the correlation-based attention mechanism in building end-to-end machine learning pipeline impact the robustness of personality prediction models using facial features? Second, can this novel feature weighting method be devised to account for non-linear interactions among facial features and their impact on personality trait prediction? and third, how can this quantification of feature weights be integrated into the model development process to ensure transparency in real-time apparent personality prediction? SVM and LR consistently demonstrate strong accuracy (86%) and F1-scores (76%), with SVM gave higher performance in neuroticism prediction.