The context
For a client in the retail sector, we built a solution to identify the most profitable locations, supporting strategic decisions on opening new stores. The solution combines advanced machine learning, data analysis and interactive visualization in an end-to-end platform.
Architecture
- Data collection and integration: data from on-prem sources is transferred and orchestrated into the Data Lake through Azure Data Factory
- Processing and training: data is prepared with Azure Databricks and machine learning models are trained on the target variables
- Prediction and storage: predictions are stored in MongoDB for fast, reliable access
- Web and mobile app: React and Node.js, with interactive maps based on the Google Maps API
Machine learning pipeline
- Demographic, economic and infrastructure data from internal and external sources (ISTAT + client data), enriched with data on the main competitors
- More than 950 features on urbanization, walkability and territorial functionality, with dimensionality reduction via PCA
- Supervised regression algorithms (Random Forest Regressor, Gradient Boosting) to predict revenue
- A profitability score for each area, summarizing the results in a single KPI
Benefits
- Strategic decisions based on objective data
- Higher returns thanks to optimal location selection
- Lower risk when opening new stores
- An intuitive interface that sales teams find easy to use
- A model that can be extended to new variables and needs



