Optimizing in-vehicle auditory systems: how music and navigation shape driver workload via machine learning

Abstract

This study investigates the effects of navigation frequency and music tempo on driver workload and driving behavior using a driving simulator. A total of 74 participants (46 males and 28 females, aged 22–40 years) were recruited. A 2 × 3 × 3 mixed design was employed, involving two navigation frequencies (high vs. low), three music tempo conditions (no music, slow tempo, and fast tempo), and three urban traffic scenarios (regular roads, school zones, and work zones). Subjective workload was evaluated using the NASA-TLX scale, while objective workload was assessed via electrodermal activity (EDA). The results indicated that slow-tempo music and high-frequency navigation prompts were significantly associated with lower workload levels. In contrast, fast-tempo music and low-frequency navigation were linked to increased workload. Notably, the interaction between fast-tempo music and low-frequency navigation significantly intensified workload, particularly in complex traffic environments such as school and work zones. Furthermore, interpretable machine learning (ML) models were developed using XGBoost and SHAP explainer, achieving over 90 % prediction accuracy in workload classification. Key predictors identified by the models included vehicle lateral position relative to the road centerline, mean driving speed, and speed variability. Distinct workload levels can be identified by specific SHAP value thresholds and particular patterns of driving behavior. These findings provide valuable insights for optimizing in-vehicle systems and developing real-time workload monitoring frameworks.