Machine learning models to identify significant factors of panic buying situation

Area of Science:

• Computer Science
• Data Science
• Behavioral Economics

Background:

• Panic buying of essential goods creates market instability and consumer access crises.
• Limited research exists on automated detection of panic-buying behavior.
• Understanding and predicting panic buying is crucial for societal stability.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting panic-buying behavior.
• To interpret classification results and identify key factors contributing to panic buying.
• To assess the performance of various classifiers on customer purchasing data.

Main Methods:

• Collected and preprocessed COVID-19 customer purchasing records.
• Applied Synthetic Minority Over-sampling Technique (SMOTE) variants and feature selection.
• Trained and evaluated state-of-the-art classifiers, including Gradient Boosting.
• Utilized explainable AI (XAI) for model interpretation and factor identification.

Main Results:

• Gradient Boosting and its advanced variants demonstrated superior performance and stability in detecting panic-buying behavior.
• Feature selection and SMOTE variants were crucial for balancing the dataset.
• Explainable AI identified key factors influencing panic-buying predictions.

Conclusions:

• Machine learning, particularly Gradient Boosting, offers a robust solution for automated panic-buying detection.
• Identifying key predictive factors can inform strategies to mitigate panic-buying crises.
• This research provides a foundation for developing proactive market stabilization tools.