Site Selection Using Machine Learning & Geospatial Analysis

📌 Overview

This project develops a data-driven site selection model to identify optimal locations for new Jollibee stores. By combining geospatial analysis, demographic data, and machine learning, the model evaluates candidate locations based on accessibility, competition, and population influence.

---

🎯 Objectives

• Identify high-potential locations for new Jollibee branches
• Analyze spatial relationships between existing stores and points of interest
• Use machine learning to rank candidate sites
• Provide interpretable insights to support business decisions

---

🧠 Methodology

1. Data Collection

• Administrative boundaries (Metro Manila / NCR)
• Existing Jollibee store locations
• Points of Interest (POIs):
  • Malls
  • Gas stations
  • Major roads / highways
• Population data

---

2. Feature Engineering

Key features include:

• Distance to nearest POIs
• Density of establishments within a radius
• Population influence per location
• Proximity to existing Jollibee stores

---

3. Spatial Analysis

• GeoDataFrames used to manage spatial datasets
• KDTree / nearest-neighbor methods for distance computation
• Radius-based counting for density estimation

---

4. Machine Learning Model

• Training and testing split applied
• Multiple models tested and refined
• Predictions used to score and rank candidate locations

---

5. Model Explainability

• SHAP (SHapley Additive exPlanations) used to interpret feature importance
• Helps identify key drivers of site suitability

---

6. Visualization

• Static maps using Matplotlib
• Interactive maps using Folium
• Final output: ranked candidate locations plotted geographically

---

📊 Results

• Ranked list of optimal locations for expansion
• Interactive map showing high-potential sites
• Insights into factors influencing successful store placement

---

🛠️ Tools & Technologies

• Python
• Pandas & NumPy
• GeoPandas
• Scikit-learn
• SHAP
• Matplotlib
• Folium

---

📌 Key Insights

• High population density strongly correlates with ideal locations
• Proximity to commercial centers (e.g., malls) improves site ranking
• Over-saturation near existing stores reduces suitability

---

🔮 Future Improvements

• Incorporate real-time traffic data
• Add rental cost / land price analysis
• Use more advanced models (e.g., XGBoost, spatial regression)
• Expand analysis beyond Metro Manila

---

🗺️ Earlier Version: Spatial Analysis in ArcGIS/Python

Before developing the machine learning model, this project was initially implemented using a purely geospatial approach in ArcGIS/Python. The goal was to identify suitable locations for Jollibee stores using spatial analysis techniques and domain-driven criteria.

🔍 Approach

The ArcGIS/Python-based workflow focused on:

• Buffer analysis around key points of interest (e.g., malls, roads, gas stations)
• Overlay analysis to combine multiple spatial layers
• Weighted suitability mapping based on predefined criteria
• Identification of high-potential zones using spatial scoring

This version relied on expert-defined rules rather than predictive modeling, serving as a strong baseline for comparison with the machine learning approach.

---

🎥 Project Demo (ArcGIS)

You can view the ArcGIS/Python workflow and results here:
👉 [https://drive.google.com/file/d/10fyYqO8W9t9ihUgFsdD2UvZ3qKXSmdCE/view?usp=sharing]

---

🖼️ Thesis Poster

A visual summary of the ArcGIS/Python-based methodology and findings:

---

🔄 Evolution to Machine Learning

The transition from ArcGIS to a machine learning approach allowed for:

• More scalable and automated site evaluation
• Data-driven feature weighting instead of manual scoring
• Improved accuracy through model training and validation
• Better insights using explainability tools (e.g., SHAP)

This progression demonstrates the shift from traditional GIS analysis to modern data science-driven decision-making.

---

👤 Author

Christian Jay P. Magsigay