Automated Personality Prediction for Military Selection and Training Analysis

Assessing personality traits is vital in military selection and operations to enhance human performance. However, relying solely on experienced psychologists for assessments introduces subjectivity and bias. With the surge in online interactions post-pandemic, there's a need to identify traits through natural interactions without psychological intervention. While machine learning (ML) algorithms have shown promise in identifying traits from social media, relying solely on them leads to accuracy issues due to imbalanced data. This paper proposes a novel two-tier oversampling strategy coupled with an ensemble deep learning (DL) method to address data imbalances. Additionally, to improve generalization capability with good accuracy, a Convolutional Neural Network (CNN) ensemble model is introduced to predict MBTI traits from natural text self-descriptions. Subject Matter Experts validated the experimental results, which show over 87% accuracy in personality trait prediction.