The main feature of modern stream applications is their strict execution time constraints. Therefore, efficiently scheduling these applications and optimizing resource usage in distributed and heterogeneous computing environments presents significant challenges. Existing scheduling solutions in edge and cloud environments has been less focus on proactive strategy to both handling the complex and dynamic nature of stream application scheduling. If exist are typically rely on either offline or online scheduling strategies. To bridge this gap, this paper introduces a Proactive Stream Task Scheduling (PSTS), a novel ML-based task scheduling method for stream applications in edge-cloud environments. PSTS is a hybrid approach, balancing offline and online scheduling by employing time series and clustering models. To deal with NP-hardness of the problem an enhancement of Gravitational Search Algorithm (GSA) version for stream task scheduling is used. Through rigorous experiments, we demonstrate the effectiveness of the proposed PSTS compared to competitive algorithms. The results show that PSTS has optimized fitness by 89%, 22%, and 19% compared to RAND, standard GSA, and GA, respectively.