Implementation Of Fuzzy-C4.5 Classification As a Decision Support For Students Choice Of Major Specialization

Determination of major specialization is important to lead the student to focus on areas of study that are of interest as well as in accordance with its academic credentials. Currently, the determination of major specialization is done by asking directly to students, regardless of academic outcomes that have been achieved in the previo us semester. This study discusses the development of a hybrid model from fuzzy Mamdani and C4.5 algorithm to analyze the determination of major specialization in informatics engineering courses of Universities Raya Serang, where C4.5 algorithm is used as a shaper rule (rule) which is used in the inference stage. Establishment of rules (decision tree) performed using Weka applications, while for the determination of the decision support analysis specialization majors using Mamdani fuzzy concept, the application is done using the help of MATLAB. The results showed that 17 of the 126 students who either choose according to variable concentrations used in this study.

💡 Research Summary

The paper addresses the problem of guiding university students in selecting an appropriate major specialization, a decision that traditionally relies on direct questioning of students without systematically incorporating their academic performance. To overcome this limitation, the authors propose a hybrid decision‑support model that combines a fuzzy Mamdani inference system with the C4.5 decision‑tree algorithm (referred to as “C45” in the implementation). The study focuses on Informatics Engineering students at Universitas Raya Serang, using a dataset of 126 individuals collected over several semesters.

Data collection and preprocessing: The authors gathered quantitative variables such as semester GPA, number of completed core courses, academic year, and gender, as well as categorical indicators of course completion. Missing values were imputed using mean substitution, and categorical attributes were one‑hot encoded.

C45 decision‑tree construction: Using the Weka environment, the C45 algorithm was applied to the preprocessed dataset. C45 selects split attributes based on the information‑gain‑ratio criterion, producing a tree that partitions students into distinct major candidates (e.g., Software, Networks, Database). Each leaf node corresponds to a clear “If‑Then” rule, such as “If GPA ≥ 3.5 and core‑courses ≥ 5 then recommend Software.”

Fuzzy Mamdani inference: The extracted C45 rules were transplanted into a Mamdani fuzzy inference system built in MATLAB’s Fuzzy Logic Toolbox. Input variables (GPA, core‑course count, year, etc.) were modeled with triangular or Gaussian membership functions labeled Low, Medium, High. The output variable represents the degree of suitability for each major candidate. The fuzzy rule base mirrors the C45 decision tree, preserving interpretability while allowing smooth handling of borderline cases through fuzzy aggregation and defuzzification (centroid method).

Experimental evaluation: The dataset was split into training (70 %) and testing (30 %) subsets. When applied to the test set, the hybrid system correctly matched the actual major choice for 17 out of 126 students (≈13.5 %). The remaining mismatches were attributed to three primary factors: (1) the limited feature set, which omitted non‑academic influences such as career aspirations, personal interests, and family expectations; (2) the manually designed membership functions and rule weights, which may not reflect optimal fuzzy parameters; and (3) potential overfitting of the C45 tree given the modest sample size.

Key contributions:

1. Integration of a classic, interpretable rule‑based learner (C45) with a fuzzy reasoning engine, achieving a balance between transparency and the ability to model uncertainty.
2. Demonstration of a low‑cost, reproducible workflow using open‑source tools (Weka for tree induction, MATLAB for fuzzy inference).
3. Provision of an initial prototype that, despite modest accuracy, validates the feasibility of data‑driven major‑selection support.

Future work suggested by the authors includes expanding the feature space to incorporate career‑goal surveys, internship experiences, and extracurricular activities; employing evolutionary algorithms or expert‑driven calibration to fine‑tune membership functions; exploring ensemble learners such as Random Forests or XGBoost to generate more robust rule sets; and developing a web‑based interface for real‑time decision support. By addressing these enhancements, the system could evolve into a reliable advisory tool for both students and academic counselors, reducing reliance on purely subjective selection methods and fostering more informed specialization pathways.