An Improved CatBoost Algorithm for Red Fox Optimization in the Field of Anomaly Detection
Hongwei Chen, Tao Wu · 2022
Anomaly detection is an important research direction for machine learning in the data era, it has been a popular technology enhancement in medica, financial and Internet fields. For anomaly detection algorithms, accuracy is an important criterion. Unbiased Boosting with Categorical features(CatBoost) is an algorithm proposed by Yandex in 2018, which claims to be superior to XGBoost and LightGBM in terms of algorithmic accuracy and other aspects. Optimization algorithms are often used instead of tedious parameter tuning when choosing hyperparameters for building models. In this paper, the Red Fox Optimization Algorithm(RFOA) is improved and great optimization results are obtained when compared with other optimization algorithms. Based on this, a CatBoost algorithm based on Improved Red Fox Optimization(IRFCB) is proposed to find the most suitable hyperparameters to build CatBoost models by training the optimization algorithm without using application scenarios. Experimental validation was conducted using anomaly detection datasets on Kaggle, and IRFCB showed good results in terms of accuracy, recall, auc, and other metrics.