Akademik Veri Yönetim Sistemi
ABANT 6TH INTERNATIONAL CONFERENCE ON CURRENT SCIENTIFIC RESEARCHES, Bolu, Türkiye, 20 - 22 Şubat 2026, ss.75-84, (Tam Metin Bildiri)
In this study, an Artificial Neural Network (ANN) based prediction model was developed to evaluate eco-driving behaviors, which are critical in terms of environmental sustainability and fuel efficiency. Using the Eco-Driving Behavior Dataset, the overall eco-driving efficiency score of drivers was predicted. Five key features were used as input parameters in the study: rpm variation, harsh braking count, idling time, fuel consumption, and acceleration smoothness. The dataset was scaled to the [0, 1] range using the min-max normalization method and divided into three subsets: 70% training, 15% validation, and 15% testing. A multilayer feedforward ANN model with a 5-8-15-8-1 architecture was designed for the study. This architecture consists of five input neurons, three hidden layers (with 8, 15, and 8 neurons, respectively), and one output neuron. Three different algorithms were used to train the model: Levenberg-Marquardt, Scaled Conjugate Gradient, and Resilient Backpropagation. The Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²) metrics were used for performance evaluation. Experimental results have shown that all three training algorithms provide high prediction accuracy. Actual and predicted values were compared on 50 randomly selected samples from the test set, demonstrating that the model can successfully predict eco-driving scores. This study demonstrates that machine learning techniques can be used as an effective tool for analyzing driving behavior and evaluating eco-friendly driving habits. The developed model is applicable in intelligent transportation systems and driver feedback systems.