Income Level Estimation with Light-GBM: Understanding Model Decisions with Explainable AI Techniques Shap and Lime
Cem Özkurt, Fatmir Garri, Bilal Emre Yahyaoğlu, Onur Ağca, Necip Furkan Bildiren, Sergen Kaynak · Artificial Intelligence in Applied Sciences · 2025
This study examines the use of machine learning and artificial intelligence algorithms to predict individuals' annual incomes. In analyses conducted using the Python programming language, the best performance was achieved in models utilizing the "Synthetic Minority Over-sampling Technique (SMOTE)" for imbalanced data sets, with an accuracy of 87.45%, precision of 85.74%, recall of 89.31%, and an F1 score of 87.30, using the "Light Gradient Boosting Machines" algorithm. Additionally, the impact of parameters and variables on income prediction was examined using interpretable artificial intelligence algorithms. The results of the study emphasize the importance of employing effective methods and explaining machine learning model predictions, as well as addressing imbalanced data sets.