Research Information System
Thesis Type: Postgraduate
Institution Of The Thesis: Gazi Üniversitesi, Fen Bilimleri Enstitüsü, Turkey
Approval Date: 2017
Student: HATİCE TIRAŞOĞLU
Supervisor: FECİR DURAN
Abstract:Cold weather and heavy winter conditions cause ice on the roads, and this is why many deaths, injuries and property accidents happen every year. In this study, an icing forecast algorithm and a mobile application were developed to prevent accidents caused by ice on the roads. With the developed application, it is aimed to give information about the formation of icing in the direction of the routes of the drivers. Temperature, dew point, real feel temperature, wind intensity, wind direction, relative humidity, wind speed and ice information from the road condition sensor and weather stations were used as training data set of the developed estimation algorithm. The training data set has two classes and indicates whether or not there is icing. After the system training is completed on the developed mobile application, forecast information about the meteorology weather features is taken and icing estimation is done for the next 12 hours. In this study, multi-layer (MLP) neural network model, linear and non-linear support vector machines (SVM) from data mining methods are used to measure the accuracy of the developed system and evaluate the data set. When the classification accuracy of the algorithms used for the performance evaluation is considered, it is the best result of MLP based on the total number of correctly classified samples. When the class in which only icing occurs is based on the correct number of correctly classified samples, Linear SVM is superior. Despite the fact that the accuracy of classification in the proposed estimation algorithm is less than that of the others, as the number of samples used in education increases, the proportion of improvement in the prediction accuracy of ice has been observed.