Driver Drowsiness Detection using Machine Learning and Deep Learning Techniques: A Systematic Review.

Area of Science:

• Computer Science
• Artificial Intelligence
• Transportation Safety

Background:

• Automated driver drowsiness detection (DDD) increasingly uses machine learning (ML) and deep learning (DL) with behavioral indicators.
• Investigating existing evidence on DDD modeling frameworks, datasets, input modalities, and performance metrics is crucial for advancing the field.

Purpose of the Study:

• To systematically review and analyze ML and DL frameworks for image- or video-based behavioral DDD.
• To compare the performance of ML and DL models in driver drowsiness detection systems.

Main Methods:

• Systematic literature search of major databases (PubMed, Scopus, Web of Science, IEEE Xplore) up to August 2025.
• Inclusion of original research applying ML/DL to behavioral features for DDD; exclusion of telemetry-only, physiological-only, reviews, and gray literature.
• Extraction of dataset details, input modalities, driving context, inference modes, ML/DL methods, performance metrics (accuracy, precision, recall, F1-score), and risk of bias assessment using PROBA-AI.

Main Results:

• 69 studies met inclusion criteria; DL models outperformed ML models in median accuracy (94.48% vs. 91.80%) and F1-score (93.15% vs. 84.00%).
• DL methods primarily used Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), or hybrid CNN-LSTM; ML methods favored Support Vector Machine (SVM) and Random Forest (RF) with handcrafted features.
• Significant methodological heterogeneity was found, with 27 studies rated as high risk of bias, impacting comparability.

Conclusions:

• Deep learning models offer superior performance for driver drowsiness detection, particularly in accuracy and F1-score.
• Methodological inconsistencies in dataset design, annotation, and validation across studies limit the comparability and generalizability of DDD systems.
• Further standardization in research practices is needed to advance the reliability and effectiveness of automated driver monitoring.