Adverse Media Classification: A New Era of Risk Management with XGBoost and Gradient Boosting Algorithms

Adverse media is negative information that is not profitable for businesses or individuals, while adverse media classification is the process of classifying news titles that are included in adverse media. In an effort to create a system capable of mitigating the occurrence of fraud for customer satisfaction, machine learning is used to classify news both as detrimental media and not for the selection of news for the customer due diligence system. This study utilizes the XGBoost and Gradient Boosting algorithms to classify news headlines. A data set of 1,281 records was collected from NewsAPI and web scraping. Back translation is used in the data preparation stage to deal with unbalanced data sets and create text variants Grid search is used to find the best hyperparameters for Gradient Boosting and XGBoost. The results of the research are in the form of a machine-learning model. Across all models examined, Gradient Boosting trained on 753 records performed best with an accuracy rate of 82.31% on test data and 84.93% on validation data. This model is able to be used to classify media and then implemented in a web-based interface.