A Comparative Study of Naive Bayes and Decision Tree Classifiers for Forecasting Attrition Risk Among AI Professionals Based on Burnout Indicators

Abstract

Employee turnover in artificial intelligence (AI)-related roles is closely linked to burnout, yet many organizations still lack a systematic way to flag at-risk staff before they resign. This study addresses that gap by building and comparing two supervised classifiers, Naive Bayes and Decision Tree, to forecast attrition risk from workplace burnout signals. A publicly available Kaggle dataset comprising 1,500 employee records across 19 attributes was cleaned through mean/mode imputation and narrowed down via feature selection before being split into training and testing subsets. Using Orange Data Mining together with 10-fold cross-validation, the two models were benchmarked twice: once on the full feature set and once after irrelevant attributes were removed. Prior to feature selection, Decision Tree slightly led with 78.8% accuracy against Naive Bayes' 77.6%; after selection, the ranking reversed, with Naive Bayes reaching 80.2% accuracy versus 80.1% for Decision Tree. The comparison suggests that neither algorithm is universally superior performance instead hinges on how the underlying features are curated, a finding that should inform how HR analytics teams design early-warning systems for burnout-driven attrition.