Detection of fatigue of vehicular driver using skin conductance and oximetry pulse: a neural network approach

Vehicular accidents are increasingly contributing towards loss of lives across the world. Timely detection of physiological and psychological parameters of the vehicular driver, which could cause various levels of physical and mental fatigue that lead to slower reflexes is therefore extremely important. As part of an ambitious research initiative, India is developing a pervasive computing solution for eliminating / reducing such accidents. As one of the component of such solution, a wearable computing system has been envisioned to be worn by the driver. A complex set of noninvasive and nonintrusive sensor-compute element integrated with appropriate e-textile would form the primary part of this wearable computer. Out of the initial set of physiological parameters such as Skin Conductance, Oximetry Pulse, Respiration, SPO2, the current work focuses on the first two parameters to detect and monitor the mental fatigue / drowsiness of a driver. Using Neural Network approach, Multilayer Perceptron Neural Networks (MLP NN) have been designed to classify Pre and Posting driving fatigue levels. The performance of single hidden layer and two hidden layers based MLP NN have been discussed using the performance measures such as, Percentage Classification Accuracy (PCLA), Mean Square Error (MSE), Normalized Mean Square Error (NMSE), Area under Receiver Operating Characteristic Curve (AROC), Area under Convex Hull of ROC (AHROC). It was discovered that the performance of one hidden layer based MLP NN is comparable to the two hidden layers based MLP NN and there is slight rise in PCLA from One hidden layer to two hidden layer.