Multimodal Personality Prediction Improving HEXACO Trait Classification Using Adaptive Attention and Deep Feature Pruning

Personality prediction has gained significant attention in affective computing, particularly in applications such as recruitment, psychological assessments, and social media profiling. Traditional personality classification methods rely on psychometric tests, which are often subjective and time-consuming. Recent advancements in deep learning have enabled automated personality assessment using facial, speech patterns and behavioral cues. However, existing models struggle with feature redundancy and computational inefficiencies, leading to suboptimal classification performance. To address these challenges, we propose a Multimodal Personality Prediction Framework that enhances HEXACO trait classification using Adaptive Attention and Deep Feature Pruning. Our approach integrates Targeted Feature Reduction Mechanism (TFRM) to eliminate irrelevant facial features and improve classification accuracy. Additionally, Adaptive Attention Fusion Networks optimize multimodal data integration, enhancing the extraction of meaningful personality traits. We evaluate our model using the ChaLearn Looking at People (ECCV 2016) dataset, which consists of 10,000 video samples labeled with Big Five personality traits. Our model applies convolutional neural networks (CNNs) for feature extraction, region-based pruning mechanisms (TFRM), and category-based mean square error (CBMSE) loss functions to refine personality trait prediction. Experimental results demonstrate that our proposed model achieves 95.9% accuracy, outperforming baseline methods. The confusion matrix analysis shows strong classification performance for extraversion and conscientiousness traits, with reduced misclassification across similar traits. Additionally, our model achieves reduced inference time (0.62s per sample) and lower computational overhead, making it suitable for real-time applications. These findings highlight the effectiveness of deep feature pruning and adaptive attention in personality assessment, paving the way for more accurate, efficient, and scalable personality classification models in real-world applications.