Implementation of the vehicle classification based-on decision tree algorithm using wireless magnetic sensors

Abstract

The design of Intelligent Transportation Systems (ITS) using wireless sensor networks to observe any road traffic, get road information, or just identify road vehicles has recently become an interesting and popular research topic because of its advantages in cost and energy efficiency. To perform this study, sensor circuit consisting of sensor node, magnetometer, power board and battery, is used. This sensor structure presents more accurate and intelligible results than sensor nodes used in other studies. Two different methods have been proposed to determine the type of vehicle with these sensor circuits. In the first method, vehicles passing by the road are classified as cars, minibuses, buses and trucks according to the proposed algorithm and 𝑀𝑆𝐿 (Magnetic Signature Length) parameters. The accuracy achieved with this method was 89%. In the other method, vehicle classification was performed using machine learning algorithm J48 which is a machine learning decision tree extension and the obtained results were optimized based on the proposed method. It uses the J48 classification algorithm implemented in Weka, a machine learning software package. The Decision Tree model was built from a series of features like magnetic raw data, measurement time derived from vehicles passing through the 3-axis HMC5983L magnetic sensor. The properties are those provided by the correct classification into the J48 training algorithm to produce a decision tree model with grading ratios that vary on the basis of cross validity. The use of J48, a machine learning algorithm, has been shown to yield more efficient and accurate results in vehicle classification. The MSL values obtained by the first method have caused difficulties in the calculation process. However, by using the J48 algorithm, more specific and sensitive boundary and threshold values were obtained. The result of the study illustrates that the vehicle classification system is so effective and efficient with an accuracy rate of about 100% with optimization of the proposed system.