AI based Psychometric Assessment using Multimodal Signal Data

This article presents a reliable AI-based psychometric assessment system using multimodal signal data such as EEG signals, speech, and structured psychometric questionnaires. Conventional psychometric testing depends on unobjective self-reporting processes without real-time flexibility. Relying on machine learning and deep learning combination, the suggested system overcomes such limitations and optimizes the precision of psychological examinations. CNN-LSTM architectures are used for emotion recognition from EEG and speech with the ability of temporal and spatial dependency extraction, and XGBoost for prediction of Big Five personality traits based on structured questionnaire responses. The approach includes modality-specific designed preprocessing pipelines, efficient feature extraction, and late fusion ensemble models for better decision-making. Data were taken from three publicly available datasets: DEAP (EEG), IEMOCAP (speech), and a psychometric dataset for the OCEAN model. Experiments yielded impressive results: 97% accuracy for EEG emotion estimation, 88.5% accuracy for speech emotion recognition, and 94.3% accuracy for the multimodal fusion model. These results demonstrate the power of multimodal fusion in psychometric assessment. The system is scalable and adaptable, suitable for use in mental health diagnosis, workplace assessment, education, and AI-augmented recruitment systems. Furthermore, it offers real-time feedback and minimizes bias, making it a promising tool for continuous psychological monitoring. Overall, this framework provides a guideline to create intelligent, ethical, and interpretable AI systems that are able to integrate physiological signals and psychological profiling within an operational environment.