Integrating Shortest Job First (SJF) Scheduling with Neural Networks for Enhanced Predictive Process Scheduling

Aditya Putra Ramdani (Department of Information Technology, Faculty of Engineering and Computer Science, Universitas Muhammdiyah Semarang, Indonesia)
Midda Restia Primadani (Department of Information Technology, Faculty of Engineering and Computer Science, Universitas Muhammdiyah Semarang, Indonesia)
Fari Katul Fikriah (Information System and Technology Program, Universitas Widya Husada Semarang, Semarang, Indonesia)
Atika Mutiarachim (Faculty of Economy and Bussiness, Universitas 17 Agustus 1945 Semarang, Semarang, Indonesia)

Journal of Computing and Smart Ecosystems
Vol. 1 No. 1 (2025): J-CaSE

Process scheduling is a critical component of operating systems, directly influencing CPU utilization and overall system efficiency. The Shortest Job First (SJF) algorithm is theoretically optimal in minimizing average waiting time but is limited by its dependence on accurate burst time estimation. This study proposes a hybrid scheduling approach that integrates neural networks (NN) with SJF to dynamically predict process execution times. The neural model was trained on process-level features, including CPU usage, memory usage, priority, and arrival time, and its predictions were embedded into the SJF mechanism. Simulation results demonstrate that the NN-enhanced SJF achieves notable reductions in average waiting time and turnaround time while improving CPU utilization compared to traditional SJF and Round Robin algorithms. These findings highlight the practical viability of lightweight predictive models for enhancing classical scheduling techniques and extend their applicability to dynamic and heterogeneous computing environments.