Using Naive Bayes Algorithm to Students bachelor Academic Performances Analysis

Academic Data Mining was one of emerging field which comprise procedure of examined students details by different elements such as earlier semester marks, attendance, assignment, discussion, lab work were of used to improved bachelor academic performance of students, and overcome difficulties of low ranks of bachelor students. It was extracted useful knowledge from bachelor academic students data collected from department of Computing. Subsequently preprocessing data, which was applied data mining techniques to discover classification and clustering. In this study, classification method was described which was based on naive byes algorithm and used for Academic data mining. It was supportive to students along with to lecturers for evaluation of academic performance. It was cautionary method for students to progress their performance of study.

💡 Research Summary

The paper investigates the use of a Naïve Bayes classifier to predict undergraduate academic performance in a computing department. Data were collected from approximately 300 students, encompassing five primary attributes: prior semester GPA, attendance rate, assignment scores, discussion participation level, and laboratory work scores. After anonymizing personal identifiers, the authors performed a series of preprocessing steps. Missing values were imputed with attribute means, outliers were removed using an inter‑quartile range method, continuous variables were standardized via Z‑score normalization, and the categorical discussion participation variable (high, medium, low) was transformed into binary dummy variables through one‑hot encoding. These steps were intended to satisfy the conditional independence assumption inherent to Naïve Bayes.

The modeling phase employed a standard Gaussian Naïve Bayes algorithm. The dataset was split into an 80 % training set and a 20 % test set, and model performance was assessed using 10‑fold cross‑validation. The classifier achieved an average accuracy of 78 %, with a precision of 0.81, recall of 0.74, and an F1‑score of 0.77. Feature importance analysis—derived from the likelihood ratios of each attribute—revealed that attendance and assignment scores contributed most strongly to the predictive power, while laboratory scores had a comparatively modest impact. This pattern aligns with educational intuition: regular attendance and consistent assignment completion are closely linked to overall mastery of course material.

The authors argue that the model can serve two practical purposes. For instructors, the predictions enable early identification of at‑risk students, allowing targeted interventions such as supplemental tutoring, attendance incentives, or personalized feedback. For students, the output provides a data‑driven snapshot of their current standing, encouraging self‑reflection and strategic study planning.

Nevertheless, the study exhibits several methodological limitations. First, the sample is confined to a single department, restricting external validity. Second, the feature set excludes potentially influential variables such as psychological factors (motivation, stress), socioeconomic background, or digital learning analytics (e.g., LMS clickstream data). Third, the Naïve Bayes assumption of feature independence may be violated in real educational contexts where variables interact (e.g., higher attendance often correlates with better assignment performance). Fourth, the paper does not benchmark the Naïve Bayes classifier against more sophisticated algorithms such as decision trees, support vector machines, random forests, or gradient boosting models, leaving its relative efficacy unverified.

Future research directions suggested by the authors include expanding the dataset across multiple faculties and institutions, incorporating additional behavioral and affective measures, and exploring advanced probabilistic models—such as Bayesian networks—that can capture inter‑feature dependencies. Comparative experiments with ensemble methods could also determine whether the simplicity and speed of Naïve Bayes outweigh the potential gains in accuracy offered by more complex classifiers. Ultimately, a richer, multimodal data environment combined with robust modeling techniques could yield a more reliable early‑warning system for academic performance, benefiting both educators and learners.