잡플래닛 데이터를 활용한 업종별 긍정 평가 예측 및 변수 중요도 분석: 군집 기반 머신러닝 접근

초록

[Purpose] This study aimed to cluster organizational perception patterns of employees in the manufacturing/chemical, service, and IT industries using corporate review data posted on JobPlanet. It also sought to build machine learning-based classification models to predict positive evaluations within each cluster. The goal was to empirically identify the presence of heterogeneous employee groups across industries and to derive key factors influencing positive organizational perception. [Methodology]The analysis consisted of unsupervised clustering and supervised binary classification. Ward linkage and K-Means clustering methods were comparatively applied, followed by classification using various machine learning models, including Gradient Boosting, Neural Network, and AdaBoost. SHAP (Shapley Additive Explanations) analysis was then conducted to quantitatively interpret the contribution of each variable. [Findings]The results showed that the optimal classification model varied by cluster, and K-Means clustering generally yielded better predictive performance across most industries and clusters than Ward linkage. SHAP analysis revealed that “recommendation of the company” was the most influential variable across all industries. In the IT sector, low ratings for “promotion opportunities” significantly contributed to negative predictions, indicating that dissatisfaction with career development opportunities adversely affected organizational evaluations. In the service sector, perceptions of “management” had the greatest impact on positive evaluations. Meanwhile, “work-life balance” had relatively low predictive power across all industries, implying that it had already been sufficiently satisfied. [Implications]This study provided an empirical framework for HR diagnosis and strategy development tailored to each cluster by integrating unstructured review texts with quantitative data and applying interpretable AI-based prediction models. The proposed methodology could help companies identify high-risk turnover groups early and develop customized HR strategies based on employee characteristics. These contributions are expected to be meaningful both theoretically and practically in the field of human resource management (HRM).

키워드