Personality Prediction Based on Video Using Transfer Learning DeepID Model

This research presents an automatic personality prediction system based on the Big Five model openness, conscientiousness, extraversion, agreeableness, and neuroticism by leveraging transfer learning on the DeepID architecture. Video input is first processed with the MTCNN algorithm for robust facial region detection under varying lighting and poses. Extracted features are fed into a modified DeepID model, pre-trained on large-scale face-recognition datasets, to perform spatial encoding. To capture temporal dynamics, Long Short-Term Memory (LSTM) networks model frame-to-frame changes in expression. Training and validation use the ChaLearn LAP dataset of approximately 10,000 annotated videos. Experimental results demonstrate 88.6% overall accuracy, with an average precision of 87.2%, recall of 86.5%, and F1-score of 86.8%, confirming the model’s balanced performance across classes. A minimum loss of 11.3% further underscores effective convergence. The complete pipeline is deployed via Flask, enabling real-time, web-based integration. Beyond academic novelty, this system holds promise for practical applications: in recruitment, it can offer unbiased, rapid personality screening; in mental-health contexts, it may assist clinicians by flagging behavioral cues non-invasively; and in human–computer interaction, adaptive interfaces could personalize responses based on users’ inferred traits. By combining transfer learning with temporal modeling, our approach delivers a scalable, non-invasive tool for automated psychological assessment through visual data, paving the way for ethical, real-time personality analytics in diverse domains.