Naive Bayes Classifier – Smoothing Method on Smartphone Sensors for Driver Activity Classification

Abstract

Most causes of death are traffic accidents. This paper aims to obtain parameters of identification of driving activities that can be developed to detect accidents in the next studies. Data is collected by sensors on smartphones, using accelerometer and gyroscope sensors. The proposed method uses Naive Bayes Classifiers (NBC) algorithm to determine driving activity, by dividing dataset into training and testing data using k-fold parameters. NBC can work using less training data, by calculating the probability value of each class from means and variance of each feature to classify classes efficiently. The results show that the accuracy of the classification is higher if a smoothing process is carried out, using single exponential smoothing method, before the clacification process of the NBC algorithm is done. The testing using 8 k-fold CV without smoothing process, using smoothing alpha (α) = 0.1, and using α = 0.9 obtain the accuracy of 98.43%, 99.27%, and 98.43%, respectively. It can be concluded that the NBC method combined with smoothing method using α = 0.1 produces greater accuracy.