Stealth assessment strategy in distributed systems using optimal deep learning with game based learning

Abstract

Game-based learning (GBL) is familiar in several respects, as it can embed problem-solving challenges in an interactive virtual environment to improve students’ motivation. Designing a stealth assessment is a tedious task that requires experienced professionals to manually derive evidence and competency approaches to precisely deduce student knowledge and skills. This necessitates that stealth assessment designers perform manual feature engineering and develop probabilistic graphical models. Since the manual feature engineering process is complex, this paper develops Optimal Deep Learning with GBL (ODL-GBL) for stealth assessment. The proposed ODL-GBL model involves parameter-tuned bi-directional long short-term memory (Bi-LSTM) with an oppositional salp swarm algorithm. The optimal Bi-LSTM model extracts hierarchical features from low-level data in an automated way (e.g., a series of student behaviours in a GBL platform) to infer evidence for competencies. The proposed ODL-GBL model alleviates the need for the costlier and more tedious process of developing evidence models. A detailed experimental analysis is used to validate the effective predictive performance of the proposed method. Among all the compared methods, the Bi-LSTM model shows superior performance, with accuracies of 60.56% and 61.34% under 70 and 40 hidden units, respectively.