.png)
1. Genetic algorithm
With the successful implementation of the apriori algorithm and the sales prediction models, it was now time to start working on the genetic algorithms, the core of this solution.
To start off, the team decided to settle on a set of heuristics to be considered by the genetic algorithms, both for the store layout and for the product placement.
It must be noted that the team wanted to keep this proof of concept as generalized as possible, without any specific store philosophies in mind. As such, the following heuristics were settled by the team as the most relevant and general ones, in which the majority of retail companies would agree:
Layout/Section planning
Hard constraints
Maximize the presence of the most wanted categories;
Minimize the distance between associated categories;
Frozen categories must be in a freezer;
Fresh categories must be in a fridge;
Categories identified for waist-level stands have to associated with the respective sections (e.g. fruits and vegetables);
Categories identified to be close to the walls have to be associated with the respective sections (e.g. butchery and bakery).
Soft constraints
Minimize the distance between similar categories (e.g. fruit and vegetables);
Minimize the distance from seasonal products to the entrance;
Categories with most products in the catalog should be placed in sections with more room.
Shelf planning
Hard constraints
Maximize the presence of the most wanted products;
Minimize the distance between associated products;
Most relevant products must be at eye level.
Soft constraints
Most relevant products should be by the middle of the corridor;
Duplicate products should be close to each other.
In the following images, a plan that the genetic algorithms came up with is presented.
By analysing the plans displayed in the figures above, the team would would say that the majority of sections seem to be in familiar or expected places, although it is difficult to evaluate, as the algorithm may have found insights that are imperceptible to users and lead to the suggested plan, despite some of the decisions seeming odd at first.
On the other hand, being a multi-objective decision making problem, it is possible that the heuristics may be too generalized and the problem requires a more complete set of heuristics in order to perform at its best, being something to look into in future work and retail companies that are looking capture their merchandising philosophy in the genetic algorithms.
2. Scientific article
Having written an overview of the solution's characteristics, benefits and limitations when compared to other existing solutions, it was now time to write a brief overview of each of the development phases undertaken during this project.
The team decided to loosely follow the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology, all the way from the business and data understanding to the implementation, evaluation and deployment phases.
With this chapter done, all that remains to be written in the scientific article is the abstract, as the team decided to leave it until everything else was written, and a few closing remarks in the conclusion, reflecting on the work that was and can be done in this subject.
3. Week retrospective
In retrospective, this was a crucial week for the project, having the longest chapter of the article to write and the core of the solution to implement.
Overall, the team sees this week's work as a success, having successfully written the development process chapter of the article with great detail and insights on the work done during the past 8 weeks, while the genetic algorithms displayed an overall impressive effectiveness when planning the store layouts and product placement, with further tweaks on the heuristics balance possibly improving its performance even further.
For the next, and final week, before the demonstrations, it is time for a final push, where a web application will be developed in order to demonstrate a real use case of this solution, the notebooks and scientific article will be reviewed before delivery, and the demonstrations will be prepared, with careful attention to the details that the team wants to present, as it would be a missed opportunity not to cover the many topics and features covered by this project and its solution.
Comments