Our research provides a comprehensive understanding of the career aspirations of computer science undergraduates based on multidimensional data that cover technical skills, soft skills, academic grades, and personal interests. To mitigate the problem of class imbalance and the presence of overlapping careers classes, categories of similar meanings were merged based on semantic similarity. This led to a more accurate definition of the target variable and streamlined the modeling process. The preprocessed data was evaluated based on numerous classification models identified as decision trees, random forests, logistic regression, K-nearest neighbors, Naive Bayes, and gradient boosting. Of the models tested, the XGBoost classifier proved to be the most efficient with an accuracy of 88% and having the potential to model the nonlinear and complex associations that exist in the data. Its performance is supported by complete classification measures and analysis of the confusion matrix that highlights its importance as a powerful analytical technique in the provision of insights on policies on education and career path formation.