Nowadays advertising is widely used to promote various products. Counting-based advertising system is a promising solution for operators to enhance system resource utilization with potential profit, which is the focus of this paper. On the system, each advertisement order comprises the display length, the contract period, and the expected play count. The objective is to fulfill the expected play count within the contract period for each advertisement. However, operators face the challenge of deciding whether they should admit new orders or not without the knowledge of total amount of available time. Therefore, we propose an efficient method based on support vector machine (SVM) to predict the available time of the system in the future. In addition, a scheduling mechanism is established in consideration of fairness among advertisements. We then conduct a series of simulations by using the real data provided by Taiwan Taxi Company to evaluate the performance of our proposed advertising system. The experiment results show that our proposed scheme can highly satisfy the needs for accepted advertisements in both prediction and scheduling phases.