Abstract— One of the predicaments of Higher Educational Institutions (HEIs) is to identify the potential schools for enrollment. Most HEIs conduct School-To-School-Promotion (STSP) to several secondary schools to sustain, if not, increases the enrollment rate. The classification techniques in data mining were used to classify feasible secondary institutions as target markets for promotion. This technique may also eliminate, if not, alleviate the expenses of HEIs by filtering which among the visited secondary schools do not contribute to the enrollment rate. Experimentation on J48, and Bayesian Network classification techniques were implemented using WEKA 3.6.0 [2] [4]. These techniques were also identified based on the accuracy of classifying the data set. C4.5 outperformed other classifying technique. The output of this research is beneficial in identifying the best classifying technique to be used in the given data set of determining which among the probable secondary schools are qualified for enrollment in the HEI. Index Terms— Bayesian Network, C4.5, classification technique, WEKA I. INTRODUCTION LASSIFICATION technique is one of the data analyses in data mining where it can be used to create models in determining the target market [1]. This technique identifies the probability of the schools to produce potential enrollees. Several classification methods were implemented and certain technique outperformed the others. This research aims to identify which among the following techniques: J48 (C4.5) and Bayesian Network works best as a classifier in the given training set of students who enrolled in the higher educational institution. Moreover, it also establishes the preciseness of the aforementioned techniques in terms of classifying instances whether the school provides enrollees or not and to determine which classifier is more accurate. The training set was used to check the correctness of classifier. Pruning was also implemented using the test set to avoid over fitting. Manuscript submitted January 02, 2014; revised January 27, 2014. This work was supported and financed by the University of the East, Manila, Philippines. S. A. Abaya is a doctoral student under the program of Doctor in Information Technology of the Technological Institute of the Philippines and currently a faculty of the Department of Computer Studies and Systems of the University of the East, Caloocan Philippines, phone: +63- 9063025931; (e-mail: sheila_abaya@yahoo.com.ph). B. D. Gerardo was with West Visayas State University in Iloilo City, Philippines. He is now the Vice President for Admin and Finance of the same institution (e-mail: bgerardo@wsu.edu.ph). B. T. Tanguilig is currently the Dean of the Graduate School of the Technological Institute of the Philippines, Quezon City Philippines (e-mail: bttanguilig_3@yahoo.com). A. Related Studies Several studies have been conducted to compare different classification techniques. Sharma, et al [3] worked on the comparative analysis of J48, ID3, ADTree, and SimpleCART classification techniques for spam emails. The research focused on the data analysis of email to identify whether the message is a spam email or not. The experiment was done using WEKA by WEKA Machine Learning Project of the University of Waikato in New Zealand. There were 4,601 instances with 1,831 spam category and 58 attributes from which 57 are continuous and 1 is nominal. The result of the experiment proved that J48 (C4.5) has the highest classification accuracy of 92.7624% where 4,268 instances were classified correctly and 333 instances were classified otherwise. Grossman, et al [10] labored on the comparison of Bayesian Network Classifier (BNC) with other algorithms of classification such as C4.5, Naïve Bayes (NB), Tree- Augmented Naïve Bayes (TAN) by Friedman et al (1997), original Bayesian network structure search algorithm (HGC) by Heckerman et al (1995), Maximum Likelihood Learners using the MDL score (ML-MDL) and two-parent nodes (ML-2P) and NB-ELR and TAN-ELR, NB and TAN with parameters optimized for conditional log likelihood of Greiner and Zhou (2002). Based on the result, BNC can be learned by maximizing conditional likelihood and thus provide a better classification probability among the other methods. II. METHODOLOGY A. Preparation of Data To identify the classification accuracy of these techniques, a training set was provided and cleaned by removing invalid data and supplying them with missing value to make sure that it provides a reliable result. The training set is actually the historical data of students who took the entrance exam and chose to enroll in a particular institution. The data considered five attributes: General Weighted Average (GWA) of secondary school students; Radial Distance, the proximity of the secondary school from the tertiary institution; School Ownership, the type of school whether publicly or privately owned; the Income Bracket, the salary range of the parents of a particular student; and the Class, it identifies whether the student enrolled or not in the organization. These criteria were categorized based on the possible value presented in Table 1. The data were stored in Comparison of Classification Techniques in Education Marketing Sheila A. Abaya, Bobby D. Gerardo, and BartolomeT. Tanguilig III C Proceedings of the International MultiConference of Engineers and Computer Scientists 2014 Vol I, IMECS 2014, March 12 - 14, 2014, Hong Kong ISBN: 978-988-19252-5-1 ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online) IMECS 2014