Model Prediksi Kebangkrutan Berbasis Machine Learning pada Sektor Ritel di Era Disrupsi Digital

Abstract

This research aims to develop a machine learning-based bankruptcy prediction model for the retail sector in the context of digital disruption. Changes in consumer behavior and increased technology-based competition increase the risk of bankruptcy, necessitating a more adaptive and accurate analytical approach. This research employs a quantitative approach with predictive analytics methods, utilizing financial data and digital variables as model input. The research stages include data preprocessing, handling class imbalance through oversampling techniques, dimensionality reduction using Principal Component Analysis, and feature selection to improve model efficiency and accuracy. Various machine learning algorithms are applied and compared, including Logistic Regression, Decision Tree, Support Vector Machine, Random Forest, XGBoost, and deep learning approaches such as Neural Networks and LSTM. Furthermore, a hybrid model is developed to optimize predictive performance by combining the advantages of various algorithms. Model evaluation is performed using accuracy, precision, recall, F1-score, and ROC-AUC metrics, and validation using cross-validation to ensure model stability and generalizability. The results show that the ensemble-based and hybrid models provide the best performance in predicting bankruptcy. The integration of digital variables has been shown to significantly improve model accuracy compared to using financial data alone. This research emphasizes the importance of a multidimensional, data-driven approach in understanding bankruptcy risk in the retail sector. The resulting model has the potential to be an effective early detection tool for companies, investors, and stakeholders in navigating business dynamics in the digital era.