Enhancing Career Counseling with Machine Learning: A Student-Centric Prediction Model
Abstract
The choice of a career is one of the most crucial career related decisions that can have a profound impact on a person's career development and career happiness. Traditional career counselling is mainly based on academic grades and psychometric tests and thus ignores broader personal and behavioural characteristics that can affect career preferences. This research introduces a machine learning framework for career prediction that takes a multidimensional profiling view based on academic and non-academic attributes, including personal interests, hobbies, language proficiency, favourite subjects and problem-solving skills, with an emphasis on the student. The student’s data and 26 predictive features that were available in a publicly accessible Kaggle dataset were used for model development and evaluation. Data preprocessing methods used were to implement robust performance assessment, class balancing, stratified 10-fold cross-validation and Grid Search-based hyper-parameter optimization. Five supervised machine learning algorithms, random forest, XGBoost, decision tree, AdaBoost and support vector machine (SVM), were tested with the metrics of accuracy, precision, and recall, and the F1 score. Experimental results showed that SVM has the highest overall performance with the accuracy of 96.37%, recall of 99.12%, and F1-score of 95.03% achieved by cross-validation. Results of confusion matrix analysis and class-wise evaluation also showed consistent predictive performance among the five career categories. A comparison of the proposed multidimensional profiling framework with the results from other career prediction studies showed the viability of the proposed framework. This result shows that the inclusion of both academic and behavioral attributes can greatly improve personalized career recommendation systems and educational counseling based on data analysis.