Optimize a chain convenience store location prediction model by using MTS-machine learning methodology.

Area of Science:

• Retail Analytics
• Machine Learning Applications
• Operations Research

Background:

• Traditional convenience store location selection is time-consuming, costly, and subjective.
• Advancements in machine learning (ML) offer opportunities for data-driven retail site selection.
• Large-scale, multi-variable location analysis presents significant computational challenges.

Purpose of the Study:

• To develop a novel hybrid approach for optimizing convenience store location selection.
• To integrate Mahalanobis-Taguchi System (MTS) for variable reduction with ML algorithms.
• To enhance data-driven decision-making in retail strategic planning.

Main Methods:

• A hybrid model combining Mahalanobis-Taguchi System (MTS) with XGBoost, Random Forest (RF), and Support Vector Machine (SVM) was developed.
• MTS was utilized for feature selection, reducing nine initial variables to five.
• The hybrid models were trained and evaluated using real-world Taiwanese convenience store location data.

Main Results:

• The Mahalanobis-Taguchi System (MTS) effectively reduced the number of predictive variables from nine to five.
• The MTS-XGBoost model demonstrated consistent prediction accuracy exceeding 75% across various training set sizes.
• MTS-XGBoost outperformed MTS-Random Forest and MTS-Support Vector Machine in predictive accuracy.

Conclusions:

• The proposed hybrid MTS-ML approach significantly improves the efficiency and accuracy of convenience store location prediction.
• This data-driven methodology reduces computational load and reliance on subjective decision-making.
• The findings offer a transformative tool for strategic planning in the convenience store retail sector.