This study investigates the relationship between collaborative communication patterns and academic performance using WhatsApp group chat data from student teams working on university projects. By applying Natural Language Processing (NLP) and statistical methods, we analyze features such as sentiment, participation equality, dialogue acts, and topic modeling to assess their predictive power for final grades. The dataset comprises approximately 10,500 messages from 72 participants across 28 student groups spanning two academic years (2023–2024). Our findings indicate that communication styles, emotional tone, and structural engagement play a significant role in team performance. Key predictors include participation levels (word count variability), sentiment balance, personality traits (particularly agreeableness and extraversion), and constructive disagreement patterns. Surprisingly, the presence of inappropriate language showed positive correlation with performance, potentially indicating intense engagement rather than disruption. The best-performing models explained 34–39% of grade variance using gradient boosting and LightGBM algorithms. While predictive power was limited by external factors and small dataset size, the results offer practical implications for educators, suggesting that balanced participation, personality-aware team formation, and sentiment monitoring could enhance digital collaboration outcomes in academic settings.