Modelling the Effectiveness of Curriculum in Educational Systems Using Bayesian Networks

In recent years, online education has been considered as one of the most widely used IT services. Researchers in this field face many challenges in the realm of Electronic learning services. Nowadays, many researchers in the field of learning and eLearning study curriculum planning, considering its complexity and the various numbers of effective parameters. The success of a curriculum is a multifaceted issue which needs analytical modelling for precise simulations of the different learning scenarios. In this paper, parameters involved in the learning process will be identified and a curriculum will be propounded. Furthermore, a Curriculum model will be proposed using the behavior of the user, based on the logs of the server. This model will estimate the success rate of the users while taking courses. Authentic Bayesian networks have been used for modelling. In order to evaluate the proposed model, the data of three consecutive semesters of 117 MS IT Students of E-Learning Center of Amirkabir University of Technology has been used. The assessment clarifies the effects of various parameters on the success of curriculum planning.

💡 Research Summary

The paper addresses the growing complexity of online education by proposing a data‑driven framework for predicting curriculum success using Bayesian Networks (BN). The authors begin by identifying the key factors that influence learning outcomes. Through a combination of literature review, surveys, and interviews with 117 master‑level IT students at the E‑Learning Center of Amirkabir University of Technology, they isolate variables across four categories: personal attributes (prior knowledge, motivation, self‑efficacy), instructional design elements (quality of materials, assignment difficulty, exam structure), behavioral indicators derived from server logs (page dwell time, click streams, video completion rates), and performance metrics (assignment scores, mid‑term/final grades, final course grade).

To capture the interdependencies among these variables, the authors select Bayesian Networks because they explicitly model conditional probabilities and can handle both discrete and discretized continuous data. Network structure learning is performed in a hybrid manner: expert‑provided constraints define permissible arcs (e.g., prior knowledge influences motivation but not vice‑versa), while data‑driven score‑based algorithms such as K2 and BDeu search for the optimal topology. The resulting network comprises 12 nodes and 15 directed edges, with Conditional Probability Tables (CPTs) estimated via maximum likelihood.

Data preprocessing involves cleaning raw server logs, converting unstructured entries into structured records, and imputing missing values using multiple imputation. Continuous variables are discretized into meaningful intervals to fit the BN framework. Model evaluation employs 10‑fold cross‑validation and Receiver Operating Characteristic (ROC) analysis, achieving an average Area Under the Curve (AUC) of 0.87, indicating strong predictive performance. Sensitivity analysis reveals that “learning motivation” and “log‑derived behavioral patterns” exert the most direct influence on success probability, while “prior knowledge” primarily acts as an indirect mediator through other variables.

The authors further conduct scenario‑based simulations to demonstrate the model’s practical utility. For instance, extending assignment deadlines or adding interactive quizzes is predicted to raise the average success probability by roughly 5 % and 8 %, respectively, whereas degrading the quality of instructional materials leads to a drop of over 10 %. These quantitative insights enable curriculum designers to allocate resources strategically, prioritize interventions that most improve learner outcomes, and anticipate the impact of proposed changes before implementation.

In conclusion, the study validates Bayesian Networks as an effective tool for modeling curriculum effectiveness in e‑learning environments. By grounding the network in real‑world log data and rigorously testing its predictive power, the authors provide a replicable methodology that can be extended to other disciplines, educational levels, or additional variables such as social‑network interactions and external resource usage. The work contributes both a theoretical framework for understanding multifaceted learning processes and a practical decision‑support system for educators and administrators seeking to optimize online curricula.