With the continuous expansion of cloud computing, traditional CPU scheduling algorithms face increasing challenges in handling complex workloads and ensuring quality of service (QoS). This paper proposes a time series–based scheduling framework, CPUSched-Forecast, which leverages multi-time series forecasting models to predict future CPU loads and integrates the predictions into scheduling decisions. To this end, we constructed task representations using a publicly available cluster dataset and designed a QoS-oriented, prediction-driven scheduling strategy that differentiates between Latency-Sensitive (LS) tasks and Best-Effort (BE) tasks. In our experiments, CPUSched-Forecast was compared with classical scheduling algorithms such as First Come First Serve (FCFS), Shortest Job First (SJF), and Round Robin (RR). The results demonstrate that our approach achieves significant improvements in key performance metrics, including Average Waiting Time (AWT), Average Turnaround Time (ATT), and CPU utilization. Notably, the SLA violation rate for LS tasks was substantially reduced, validating the effectiveness of prediction-driven scheduling in enhancing system performance and ensuring QoS. This study highlights that integrating time series forecasting with scheduling strategies provides an effective approach to improving resource utilization and service quality in cloud computing, offering practical insights for the development of intelligent and efficient resource management frameworks.