Grid search vs Bayesian optimization for intensity scoring classification and channel recommendation prediction

Abstract

Technological advancement has spurred financial technology growth, transforming traditional financial operations into digital. Peer-to-peer (P2P) lending is a key fintech solution offering online loans, though it struggles with repayment issues due to customer financial instability. To overcome these challenges, XYZ is a startup that focuses on enhancing the efficiency of collections and communication with customers. XYZ necessitates the implementation of a collection intensity scoring (CIS) model and a prediction model for interaction on recommended communication channels in order to optimize the collection process. This study evaluates the performance of grid search and Bayesian optimization on random forest (RF) classification models and K-nearest neighbors (KNN) regressor prediction models. RF and KNN regressor algorithms optimization are necessary to enhance their performance in CIS classification and channel recommendation prediction. The research stages follow the cross industry standard process-data mining (CRISP-DM) framework, which consists of business understanding, data understanding, data preparation, modeling, and evaluation. The model performance is assessed by accuracy and mean absolute error (MAE). The results of this study show that Bayesian optimization surpasses grid search, enhancing the accuracy of the RF model to 98.34% and reducing the MAE of the KNN regressor model to 0.24530.