Exploring the Relationship between Career Satisfaction and University Learning Using Data Science Models

Abstract

Current research on the career satisfaction of graduates limits educational institutions in devising methods to attain high career satisfaction. Thus, this study aims to use data science models to understand and predict career satisfaction based on information collected from surveys of university alumni. Five machine learning (ML) algorithms were used for data analysis, including the decision tree, random forest, gradient boosting, support vector machine, and neural network models. To achieve optimal prediction performance, we utilized the Bayesian optimization method to fine-tune the parameters of the five ML algorithms. The five ML models were compared with logistic and ordinal regression. Then, to extract the most important features of the best predictive model, we employed the SHapley Additive exPlanations (SHAP), a novel methodology for extracting the significant features in ML. The results indicated that gradient boosting is a marginally superior predictive model, with 2–3% higher accuracy and area under the receiver operating characteristic curve (AUC) compared to logistic and ordinal regression. Interestingly, concerning low career satisfaction, those with the worst scores for the phrase “how frequently applied knowledge, skills, or technological tools from the academic training” were less satisfied with their careers. To summarize, career satisfaction is related to academic training, alumni satisfaction, employment status, published articles or books, and other factors.