Driver Drowsiness Detection: A Machine Learning Approach on Skin Conductance

Area of Science:

• Biomedical Engineering
• Transportation Safety
• Wearable Technology

Background:

• Drowsy driving is a major cause of global car accidents.
• Drivers may not recognize their own drowsiness, necessitating objective detection methods.
• Previous drowsiness detection systems were often large and intrusive.

Purpose of the Study:

• To investigate the efficacy of a single, comfortable wrist-worn device for detecting driver drowsiness.
• To analyze the physiological skin conductance (SC) signal for drowsiness indicators.
• To compare the performance of different ensemble algorithms for drowsiness detection.

Main Methods:

• Utilized a single wrist-worn device to collect skin conductance (SC) data.
• Applied signal processing techniques to the SC data.
• Tested three ensemble algorithms, including Boosting, to classify drowsiness states.

Main Results:

• The Boosting algorithm achieved the highest accuracy (89.4%) in detecting driver drowsiness.
• Drowsiness detection was feasible using only the skin conductance signal from the wrist.
• The study demonstrates the potential of a non-intrusive, single-sensor approach.

Conclusions:

• A single wrist-worn device analyzing skin conductance is a viable method for detecting driver drowsiness.
• This approach offers a comfortable and less intrusive alternative to existing systems.
• Further research is warranted to develop a real-time drowsiness warning system.