Rescheduling Exams Within the Announced Tenure Using Reinforcement Learning

Abstract

The academic examinations are scheduled as per the academic calendar. However, due to the occurrence of unprecedented events the exam schedule will get disturbed which in turn impacts the academic calendar. In this paper, we propose a temporally optimal and topic complexity balanced re-schedule model in terms of a Markov Decision Process (MDP). This MDP is verified for the feasibility of a schedule using Bellman Equation with Temporal Difference Learning, policy iteration and value iteration algorithms of Reinforcement Learning. The objective of the proposed model is to find an optimal mapping from the state of disturbed exam to the state of plausible date of the exam. The novelty of this work is manifested in variable penalties assigned to various schedule slippage scenarios. Rescheduling is optimally automated using Bellman Equation with Temporal Difference Learning. These variable policy iteration and value iteration algorithms have demonstrated that as the learning progresses the optimal schedule gets evolved. Also, it is shown that policy iteration has converged faster than the value iteration while generating schedule. This instils confidence in utilizing the Reinforcement Learning algorithms on the grid environment-based exam rescheduling problems.