Human behavior is complex and often resists straightforward modeling with traditional mathematical approaches. To simplify this task, researchers frequently employ intermediate psychological models that capture specific facets of human behavior. These models, like the Big Five personality framework, are typically validated using survey instruments and are known to correlate with certain behavioral tendencies. Traditionally, these constructs have been used to predict stylized behaviors; however, advances in sensing technologies have opened up new possibilities to infer these psychological constructs directly from observed behavior.
Modern smartphones are equipped with a variety of sensors that can be leveraged to capture abstract measures of human behavior. This raises the question: can we reliably infer psychological profiles from passive smartphone data alone? The ability to derive a personality profile from unobtrusive, sensor-derived data has promising applications, ranging from personalized marketing to targeted social or health interventions.
In this study, we developed a model to infer personality traits based on the Big Five personality inventory. By analyzing daily routines captured via smartphone sensors, we applied supervised machine learning to predict individuals’ personality traits. Our evaluation, using cross-validation, showed that the model achieved a sufficiently low root mean squared error to provide actionable predictions for most individuals, though it struggled with personality outliers.
This project demonstrates the feasibility of using mobile sensor data to approximate personality traits, suggesting potential for real-world applications in fields requiring adaptive and personalized approaches.