The use of new information and communication technologies is an important aid to solving transportation problems. This is commonly known as ITS (Intelligent Transport Systems) that can provide effective information for travelers and traffic managers. Road traffic prediction has led to a growing research area debated by several researchers affiliated to a range of disciplines. Recently, a significant amount of research efforts has been devoted to deep learning methods, greatly advancing traffic prediction abilities. The purpose of this paper is to meet a practical need: forecasting highway traffic volume by payment method (manual/ electronic toll) over a long time horizon based on historical observations. A variety of methods were used to make this forecast, and the results were compared and interpreted to identify model limitations and improve the approach. We propose to forecast daily highway traffic using four methods: Seasonal Autoregressive Integrated Moving Average with additional lagged values (SARIMAX), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) and at least we perform combination method which has been shown to be efficient in the literature and which, applied to our real case study, will confirm these performances. Indeed, the best prediction results were obtained using the hybridization CNN-LSTM which reached an R-squared value of 95%. In the present study, such an effort resulted in considerable improvement in long terme forecasting accuracy when compared with the existing models' performance previously adopted in the field of traffic forecasting.