Predicting Potential Deposit Customers Using CatBoost with Feature Selection and Hyperparameter Optimization

Abstract

The banking industry requires increasingly targeted marketing strategies to identify potential customers for deposit products. This study aims to build a predictive model for identifying potential deposit customers using the CatBoost algorithm combined with feature selection based on feature importance and hyperparameter tuning using Hyperopt. This study employs an experimental method using the Bank Marketing Dataset from Kaggle, a secondary dataset consisting of 45,211 data points and 17 attributes, including a target variable indicating the customer’s decision to subscribe to a deposit product or not. The research stages included Exploratory Data Analysis (EDA), preprocessing, data cleaning, class balancing using SMOTE-NC, feature selection, hyperparameter tuning, modeling, and model evaluation. The EDA results revealed outliers in the duration and previous features, as well as class imbalance in the target variable. After preprocessing, the model was built using 10 selected features: duration, month, contact, day, outcome, balance, housing, pdays, age, and job. Evaluation results using a confusion matrix showed that the CatBoost model achieved an accuracy of 92.8%, a sensitivity of 91.0%, and a specificity of 94.8%. These findings demonstrate that the combination of CatBoost, feature selection, and hyperparameter tuning is capable of producing an accurate and efficient predictive model for handling both numerical and categorical data. This model can help banks develop more effective, cost-efficient, measurable, and data-driven deposit marketing strategies, while also supporting business decision-making that is more adaptive to customer behavior.