DATA STORIES | CLUSTERING | KNIME ANALYTICS PLATFORM
Retail Store Clustering with just one click
Use clustering algorithms to understand market demand and improve your competitive advantage
As first published in BI-FI Blogs
Retail store clustering is a data analysis technique that groups similar stores based on various characteristics. It allows retailers to identify regional market trends and preferences, enabling them to tailor their product offerings and marketing strategies to meet local demand. The key advantages are listed below:
- Better understanding of market demands.
- Improved store planning and operations.
- Better resource allocation.
- Improved customer experience.
- Competitive advantage.
With our clustering solution, you can customize the inputs, assign specific weights to each input, analyze the results, and iterate the process as many times as you wish until you get a satisfactory result.
In this article, we’ll go over our Clustering Solution and what users can do to use their own data. You can download the workflow “Retail Store Clustering” for free from the KNIME Community Hub.
The sample data used for this workflow can be downloaded here.
First, the user should run the 1. STEP and fill the input sections with the data file location, cluster count and weight for each feature used in the model.
In the 2. STEP, the data is loaded in KNIME Analytics Platform and different data sheets are joined together. If you are using a different dataset, it is important to keep the same sheet names as in the sample data since the sheet names are passed as a parameter to import the corresponding sheet. Alternatively, make sure to adjust the parameters to import sheets with different names.
Then in 3. STEP, you have the option to exclude certain stores from the model because they might be outliers in terms of their features.
In the 4. STEP, you might want to cluster the stores within their respective capacity group. For example, filtering only big stores will cluster only those stores and exclude others from the model. This might be useful if your allocation plans is highly dependent on the store size.
In the 5. STEP, user-defined weights are applied to features and several transformation steps are performed (e.g., string to number conversion, normalization, etc.).
To assign weights to each unknown number of product category column, we use the Column List Loop Start and Loop End (Colum Append) nodes.
These loop nodes (together with a few manipulation nodes) allow us to rename each category column to anonymous column and apply the user weight to each column. With these nodes, no matter how many columns of product category you have, you will be able to assign weight to each of them.
Note. There might be some adjustments needed in some nodes if your dataset has more features than the sample dataset. Please leave a comment or reach out to us if that’s the case.
The preprocessed data table can now be sent to 1. and 2. model (6. STEP) where we implement in parallel a k-Means and a Hierarchical Clustering model using KNIME’s native analytics nodes.
In the 7. STEP, the groups identified by the clustering algorithms are used to create a dashboards with results.
The result dashboards contains:
- Cluster store count distribution
- Store count by cluster and store capacity group
- Cluster — store details
- Cluster averages for each feature
Additionally, you can see the cluster store distribution on a map with the help of OSM Map View node. You can customize the map tooltip by using the Column Filter node in the upper branch inside the 1. or 2. model metanodes.
Finally, in the 8. STEP, you can also compare the performances of the two models by checking their Silhouette Coefficients.
You can also export the results to Excel with the Excel Writer node for further analysis.
In conclusion, retail store clustering is an essential tool for retailers to gain a competitive advantage and improve their overall performance. With our solution, you can easily cluster your stores with just a few clicks.
If you liked this article, please don’t forget to share it and leave a comment below!
