INDONESIAN FRESH GRADUATES EMPLOYABILITY CLASSIFICATION USING DATA MINING TECHNIQUES

• Main Author: Latifa, Laili
• Format: Theses
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/80561
• Institution: Institut Teknologi Bandung

The role of universities in preparing graduates to be ready to face the world of work has always been a controversial issue, universities are often criticized for the unpreparedness of their graduates to be involved in real contexts in professional practice, in which case the employability of graduates is one of the main problems that occur in higher education. Therefore, universities are increasingly focused on preparing their graduates to be able to get jobs after graduating from college. Universities and the Indonesian government have implemented various programs to improve the quality of both universities and their graduates so that they are ready to face the world of work and can reduce unemployment. One of the programs carried out by the Ministry of Education, Culture, Research and Technology (Kemdikbudristek) is the construction of a tracer study system that aims to track the activities of graduates after a period of higher education. There has been no research that has developed a model that is used to predict employability of university graduates in Indonesia using datasets derived from the Indonesian tracer study database. Therefore, this study aims to build an employability classification model for university graduates in Indonesia using data mining techniques that can predict how graduates will work after graduating from college. The employability classification model consists of a classification model for predicting the employment status of university graduates within six (6) months after graduation and a classification model for predicting the match between field of study and work obtained; and also investigate important variables that are relevant to the model. This study uses data sources derived from Belmawa’s tracer study database, amounting to 365.062 graduates and ITB’s tracer study database, amounting to 8.274 graduates. The employment status classification model based on the Belmawa’s tracer study dataset shows that the model built using the Random Forest - SMOTE-ENN algorithm gives the best performing model with an accuracy of 95.67% and F1 of 95%, with 10 important and relevant variables consisting of number of companies applied for, grouping of college supervisors, time to look for work, fields of study, looking for work through relationships, looking for work via the internet, GPA, emphasis on learning aspects of internships, looking for work through networks, and English competency. In the employment status classification model based on the ITB’s tracer study dataset, the Random Forest - SMOTE-ENN is also the algorithm that provides the best performing model with an accuracy of 94.45% and F1 of 94%; 10 important and relevant variables consist of time looking for work, number of companies applied for, looking for work via the internet, fields of study, sources of tuition funds, looking for work through relationships, GPA, looking for work through job fairs, foreign language competence, and emphasis on learning aspects discussion. The second model that is built is a classification model of the match between field of study and work obtained by college graduates. The work match classification model built using Belmawa’s tracer study dataset shows that the Random Forest - SMOTEENN algorithm provides the best performance model with an accuracy of 96.75% and F1 of 67%, where 10 important and relevant variables consist of education level with occupation, grouping of college supervisors, type of company, fields of study, discipline-knowledge competency, monthly income, emphasis on learning in the research aspect, GPA, emphasis on learning in the discussion aspect, emphasis on learning in the practical aspect, and level of study. The work match classification model that is built using ITB’s tracer study dataset with the Random Forest algorithm produces a model with an accuracy of 99.35% and F1 of 99% with important and relevant variables consisting of sources of tuition funds, discipline-knowledge competency, type of company, monthly income, outside discipline-knowledge competency, and fields of study.