Performance Analysis of Driver Abnormal Behavior Classification System Using Enhanced Multi-Layer Perceptron

Traffic accident is one of the contributing factors of the death that happened around the world. From various types of drivers that are usually found in the traffic, motorcycle drivers have a relatively higher risk compared with other types of drivers. Therefore, there is a need for a system that is capable to detect dangerous behavior from the driver and gives an alert when abnormal behavior is detected. To develop it, we choose the 2 basic sensors to detect the movement from a driver, which are Accelerometer and Gyroscope. These sensors have been integrated into a smartphone. For the classification process, we proposed an Enhanced Multi-Layer Perceptron (MLP). The complexity of the model is reduced to ensure that our proposed system will be able to work in real-time conditions and in a limited-resources environment. In our model, we use the combination of ReLU and Softmax function, two of the famous Activation function, to enhance the performance of our model. The level of accuracy of our model achieved 97.5% with an average computational time of 45 ms. This proved that our model works better than the previous research with the same dataset.