In a paper published in the journal Scientific Reports, researchers have presented a pioneering study addressing the escalating turnover trend among new college graduates. Their innovative approach involves the development of a machine learning (ML)-based turnover intention prediction model, utilizing data from the Korea employment information service's job movement path survey.
Study: Machine Learning Insights for New College Graduate Retention. Image credit: Ground Picture /Shutterstock
Notably, the study reveals a shift in the impact of traditional factors on turnover intention, with job security emerging as the foremost predictor. The model achieves remarkable accuracy by employing logistic regression (LR), k-nearest neighbor (KNN), and extreme gradient boosting (XGB) classifiers, signaling a significant advancement beyond conventional econometric models. This research offers nuanced insights into factors influencing turnover intentions, providing valuable guidance for organizations in effectively managing and retaining emerging talent while highlighting the evolving relevance of traditional considerations.
Related Work
Past research in Korea has highlighted the pressing issue of youth employment, especially the challenges college graduates face, leading to a societal concern. The frequent turnover of new employees, particularly college graduates, has been identified as contributing to the need for more human resources in companies. The study underscores the need to swiftly classify new employees with turnover intentions, a complex task given the sensitivity of the issue and the concealment of such intentions. Previous studies using traditional econometric models have faced limitations in predicting turnover intentions.
Analyzing Turnover: Methods and Findings
This study analyzes factors influencing turnover intention among new college graduates and proposes a predictive model for turnover intention. The study acknowledges the potential for omitted variable bias in previous research and addresses this concern by treating each specific questionnaire item as an independent variable. Aligned with existence, relatedness, and growth (ERG) theory, job choice motivation is categorized into existence, relatedness, and growth needs. Considering influencing factors, the study includes personal job fit and job preferences.
Researchers addressed potential concerns about standard method bias by conducting a principal component analysis (PCA) through scientific kit-learn's (scikit-learn) PCA class to execute Harman's single-factor test. Results indicated that the first principal component accounted for 31% of the data variance, below the commonly used threshold of 50%, confirming the absence of significant standard method bias.
The study utilized data from the 2019 graduates occupation mobility survey (GOMS), comprising 18,163 sample individuals who graduated within the past two years. Researchers predicted turnover intention among employed respondents by selecting 12,202 samples, where 76.54% indicated 'no' and 23.46% indicated 'yes' for turnover intention.
Cronbach's Alpha values for the variables used were 0.807 or higher, confirming reliability. Researchers established discriminant validity by ensuring the most negligible average variance extracted (AVE) squared value exceeded the most significant correlation coefficient among concepts, meeting criteria requirements.
Ordinary least squares (OLS) regression analyzed path coefficient values and significance levels. ML techniques, including LR, KNN, and XGB, were employed to construct a predictive model for turnover intention. Highlighting the advantages of ML, researchers emphasized its ability to handle both categorical and numerical predictors and assess linear and non-linear relationships. The study's robust approach integrates theoretical frameworks, addresses methodological concerns, and leverages advanced analytical techniques to understand and predict turnover intention among new college graduates.
ML Insights on Turnover
The study conducted a regression analysis to explore the impact of various independent variables on turnover intention, revealing that job preference variables, except for satisfaction with work's social reputation, significantly influenced turnover intention. The subsequent analysis of the predictive models indicated that XGB demonstrated the highest accuracy, with 78.5%, followed closely by LR at 78.3%, and KNN, with the lowest accuracy, at 76.1%. Further investigation into the importance of features highlighted the critical role of satisfaction with job security as the most significant predictor of turnover intention.
The findings also delved into the nuanced influence of job preferences, uncovering that while recognizing workload importance, recognizing the importance of the significant field, and recognizing work's social reputation before employment did not impact turnover intention, post-employment satisfaction with social reputation was associated with increased turnover consideration. Additionally, personal-job fit, including alignment with one's primary and satisfaction with individual development potential, emerged as pivotal factors influencing turnover intention.
The study's innovative integration of ML techniques and traditional econometric analysis provides a comprehensive understanding of turnover intention among new college graduates. Notably, it challenges conventional assumptions about the impact of certain variables, such as job preferences, and establishes a hierarchy of critical predictors. The practical applications of the study extend to organizational strategies, emphasizing the importance of addressing job security and organizational culture and aligning job roles with individual skills to reduce turnover intention effectively.
Despite its contributions, the study acknowledges limitations, such as using cross-sectional public data, cautioning against definitive causal inferences. Validation across diverse populations and addressing data imbalances in turnover intention offer avenues for future research enhancement.
Conclusion
To sum up, this study combines traditional econometric analysis and ML to delve into turnover intentions among new college graduates. Job preference variables significantly impact turnover intention, with job security satisfaction emerging as crucial. XGB demonstrates the highest accuracy among predictive models.
The findings challenge conventional assumptions about certain variables, emphasizing the nuanced influence of job preferences. Practical implications stress organizational strategies targeting job security and a positive culture to mitigate turnover. While acknowledging limitations, the study advances understanding and offers actionable insights for organizations managing new talent.
Machine Learning Optimizes Polymer Analysis