Footfall and Sales Analysis Dashboard: Uncovering Hidden Relationships in Retail Data with Explainable AI

This project is an interactive retail analysis dashboard built on Streamlit. It combines linear regression prediction and SHAP explainable AI technology to deeply analyze the relationship between footfall and weekly sales at Walmart stores, providing data support for retail decision-making. The project emphasizes interpretability, helping managers understand the business logic behind the data instead of relying on black-box models.

In the retail industry, the relationship between footfall and sales is complex: Does the peak footfall during holidays translate into sales growth? How does temperature affect consumer behavior? What are the patterns behind performance differences across stores? The Footfall-Sales-Dashboard project was created to answer these questions. It integrates machine learning prediction and explainable AI analysis to help practitioners uncover insights from data.

The project uses a Python toolchain to build an end-to-end analysis process:

• Streamlit: Quickly build interactive web interfaces, simplifying front-end development;
• Pandas: Handle data preprocessing such as cleaning and transformation;
• Scikit-learn: Use linear regression models for sales prediction;
• Matplotlib/Seaborn: Implement data visualization;
• SHAP: Explain model decision logic based on game theory's Shapley values.

The project's core features include:

1. Interactive dashboard: Support data filtering, drilling down, and comparison;
2. Linear regression prediction: Use classic algorithms to predict sales with transparent and interpretable coefficients;
3. Footfall and sales analysis: Identify conversion efficiency and seasonal patterns;
4. Correlation heatmap: Display relationships between variables;
5. Holiday/temperature impact analysis: Optimize promotions and inventory allocation;
6. Store-level insights: Analyze performance differences by store dimension;
7. Interactive filters: Dynamically filter data to focus on specific scenarios.

The project's biggest feature is the integration of SHAP technology:

• Solve the black-box model problem: Not only provide prediction results but also explain the contribution of each feature (e.g., footfall, temperature, holidays) to the results;
• Working principle: Predicted value = baseline value + sum of SHAP values of each feature, quantifying the direction and magnitude of each feature's impact;
• Application value: Help managers identify key factors, understand the causes of sales fluctuations, and build trust in the model.

The project applies to various retail scenarios:

• Store operation optimization: Identify low-conversion stores and develop improvement measures;
• Promotion effect evaluation: Quantify the contribution of holiday promotions;
• Inventory management: Optimize inventory based on predictions;
• Staff scheduling: Allocate staff according to footfall patterns;
• Strategic decision-making: Provide data basis for store opening/renovation.

The project can be improved in the following directions:

1. Model complexity: Try XGBoost/LightGBM to improve accuracy while maintaining SHAP interpretability;
2. Data dimensions: Introduce more features such as weather and competitor activities;
3. Real-time performance: Extend to real-time data stream processing to support dynamic monitoring;
4. Multi-store comparison: Enhance cross-store comparison analysis to support benchmarking learning.