A hybrid genetic algorithm-neural network framework for adaptive university timetable scheduling

Abstract

Timetable scheduling in higher educational institutions is a complex NP-hard optimization problem that requires the allocation of courses, lecturers, classrooms, and time slots while satisfying numerous hard and soft constraints. Traditional manual scheduling methods are time-consuming, error-prone, and difficult to adapt to dynamic institutional changes. Although automated techniques such as Genetic Algorithms (GA) have been widely used to address this problem, they often suffer from slow convergence, high computational cost, and limited adaptability to evolving scheduling patterns. This study proposes a hybrid Genetic Algorithm–Neural Network (GA–NN) framework that integrates the global optimization capability of Genetic Algorithms with the predictive learning ability of Neural Networks. The proposed system was implemented in Python using the DEAP evolutionary computation framework and TensorFlow for neural modeling, and deployed as a scalable web-based platform using Django and PostgreSQL. In the hybrid model, the Neural Network assists the evolutionary process by guiding mutation operations and improving the fitness evaluation of candidate schedules through predictive quality estimation. Experimental evaluation conducted on multiple dataset scales demonstrates that the proposed model significantly enhances scheduling performance. Results show a 78.5% reduction in hard constraint violations, 39.5% reduction in computation time, and approximately 42.7% faster convergence compared with a standard Genetic Algorithm. Furthermore, the system achieved 91.2% soft-constraint satisfaction and a 4.7/5 user satisfaction rating, demonstrating its effectiveness, adaptability, and practical applicability for modern university timetabling.