Recently, the rise of used cars sales is exponentially increasing more than ever. An accurate market price prediction is important for not only buyers but also dealers to get more profit in a tight race. The approaches commonly used for price prediction task and the other similar problems are linear regression analysis, decision tree or fully connected neural network. These approaches are popular due to its easy implementation; however, the low prediction accuracy
is a massive drawback. In this paper, a novel Deep Learning framework is proposed to overcome this problem. From experimental results, the relative error was found out to be 8.18% that is about 4.47% smaller than the other conventional Machine Learning techniques. The evidence indicates that our proposed framework has great advantages compared to the existing ones, as it not only gives more accurate predictions but also considers the prediction uncertainty. It is important because it assesses how much to trust the forecast produced by the model. The whole experiment is based upon real-world data of Korean used cars.