The study started with a realistic need to contribute to enhancing the future competitiveness of universities by providing guidelines and supporting career development through precisely predicting career direction for getting talented students in a ra...
The study started with a realistic need to contribute to enhancing the future competitiveness of universities by providing guidelines and supporting career development through precisely predicting career direction for getting talented students in a rapidly changing modern society. Based on those purposes, the study focused on developing a predictive model for classifying the young adults’ career status and finding out which factors affect most importantly the young adults’ career status by analyzing the characteristics of major predictive factors.
The data based on the purpose of the study, the 2005 Korean Education Longitudinal Study(2005KELS) by the Korea Education Development Institute(KEDI), were used. Before starting analysis, the group was divided into two groups: groups that graduate from high school(g1: job seekers or employed, g2: college students) and groups that graduate from college(g1: job seekers or the employed, g2: graduate students) in accordance with the characteristics of the data. Next, to find out which variables affect the prediction of the youth’s career status through creating a prediction model, not only were classic statistical models created using logistic regression and discriminant analysis, but machine learning models were also made using the Decision tree, Support Vector Machine(SVM), Random forest, and XGboost algorithms.
In a consequence of the analysis, the SVM prediction models turned out to be the best prediction rates in both groups 95.45%(high school graduation groups) and 82.12%(college graduation groups), respectively. Therefore, the final predictive model was decided to SVM model, and it was analyzed to the main predictive variables from the SVM model. Thus, the top predictive variable from the high school graduation groups was the number of friends contacting their acquaintances frequently belonging to personal environmental factors, and the number of people who could call for the help belonging to personal environmental factors was the top predictive variable from the college graduation groups.
Next, the top 10 major predictors were examined for each group as factor categories. As a result, it was found that personal environmental factors, personal feature factors, and SES(Social Environment Status) factors were the main predictors of the young adults’ career status in the high school graduation groups. However, it is necessary to pay more attention to personal environmental factors and personal feature factors(especially social relationships) than SES factors because they are the third-ranked factors and have already been verified many times in lots of previous studies.
In the college graduation groups, personal feature factors were the key factors predicting the young adults’ career status. Therefore, it is necessary to examine how personal feature factors(intelligence, values, competencies) affect their careers in relation to the competencies currently being established as policies in many universities. Furthermore, the importance of social relationships in the young adults’ careers needs to be reexamined as social relationships were the key factors among individual environmental factors in all the machine learning models.
Lastly, the results about all predictive models analyzed in this study are summarized as follows: First, the best fitting model was machine learning models rather than classic statistical models in order to predict the young adults’ career status in the high school and college graduation groups based on their educational level. Therefore, it is necessary to examine various variables that have not yet been put into the machine learning models to identify various variables that affect the young adults’ career status.
Second, it still needs to be used in both classic statistical methods and machine learning models. If the volume of data is not as many as creating machine learning models, it is recommended to use classic statistical methods as a prototype model. If significant factors are not yet been found in areas where there have not many prior studies and the volume of data is large, it is recommended to create a prediction model using machine learning.
Third, when it is needed to use machine learning, it is necessary to selectively choose appropriate algorithms depending on the purpose and situation. First of all, even if the prediction rate is somewhat low, the decision tree model can be useful and fit into some areas where the decision-making process is more important than corrects prediction. On the other hand, in situations where accurate prediction is the most important, such as in the medical field, the SVM model is the most suitable. However, it needs to be accompanied by machine learning knowledge and empirical knowledge because it is required to set the classification function(kernel) adjustments and other tuning processes. Meanwhile, the Random forest model is recommended in a situation where decision-making is required through proper visualization if prediction rate is important and also easy to handle. On the other hand, XGboost is recommended rather than the Random forest model for highly-skilled users who can make precise and delicate tuning adjustments in machine learning.
Finally, if the prediction model is used for high school or college students, it is possible to predict who is more likely to go to university or graduate school as practical parts. Therefore, it can be helpful to design useful guidelines for them to enter higher educational institutions. Also if the predictive model is used for students who are not in a career decision-status college students, their future career status can be predicted. Based on this, it can be presented by personalized career design guidelines along with explanations and advice on expected careers after graduation. Through this, it is expected to contribute to enhancing the competitiveness of the university.