Cloud computing is a fundamental paradigm for computing services based on the elasticity attribute, in which available resources are effectively adjusted for changing workloads over time. A critical challenge in such systems is the task scheduling problem, which aims to identify the optimal allocation of resources to maximize performance and minimize response times. To overcome these drawbacks, a novel latency‐aware time series‐based scheduling (LATS) algorithm has been proposed in this paper for predicting future server loads. The proposed method involves collecting workloads, preprocessing and clustering them, predicting time series, and post‐processing the data. The workload data will be divided according to a historical time window during the preprocessing phase. Next, the time series data will be clustered based on the latency classes using the dynamic fuzzy c‐means algorithm. The time series prediction phase utilizes the Gated Recurrent Unit (GRU), and post‐processing is performed to retrieve the original data. An evaluation of the accuracy of future workload predictions was conducted based on actual requests to web servers, and the silhouette score was utilized as the metric for assessing cluster performance. The proposed model has been compared with previous approaches involving Crystal LP, SWDF, and GA‐PSO approaches in terms of prediction accuracy by 31.9%, 18.74%, and 12.16%, respectively.