An Resignation Prediction Method Based on KTE-SMOTE

Aiming at the data imbalance problem in employee resignation prediction, this paper proposes a method based on the improved SMOTE algorithm. The data imbalance problem is one of the main challenges in employee resignation prediction, and the traditional SMOTE algorithm often suffers from sample generation bias and over-generation of samples in the central region when dealing with imbalanced data, which leads to poor performance of the prediction model. In order to solve these problems, this paper proposes a KTE-SMOTE algorithm, which improves the stability of the algorithm by introducing the k-means algorithm, selecting the optimal number of clusters k by using Silhouette Coefficient, combining with under-sampling of Tomek Links, and smoothing the boundary when generating samples by edge smoothing technique. Finally, ENN is used for data cleaning to ensure the diversity of generated samples and the clarity of boundary samples. The experimental results show that the KTE-SMOTE algorithm exhibits better performance than the traditional SMOTE algorithm in a variety of machine learning models.