BrainStation Capstone Project - Grocery List Prediction + Recommender System

This page includes:

• Problem Statement
• Order of reading the notebooks
• Source of initial data

Problem Statement:

For years large grocery chains were offering membership(loyalty) cards with a point system to their customers. Idea is simple, customers would sign up and show the card every time they shop and for each dollar spent (on selected items) they would receive approximately 10 points. Once customers accumulate enough points, they are able to use them in order to reduce the total dollar amount for their next bill. Unfortunately, people have to spend in the neighbourhood of 700 dollars, overall, to accumulate enough points for a 10 dollar discount. On the other side of the process, supermarket chains receive very useful data, aggregated across all members and all stores. The collected data is then used for inventory management, product selection, targeted discounts, bundle offers, supply chain management and more. In other words, with the collected data companies are able to save money by cost reductions and increase profits by knowing which products/bundles to promote for higher sales. However, is there a way to alter the strategy in a way that will bring additional value to both parties?

What if, in addition to the point system, customers would have no need to remember which products are running low at home?! With having enough data and sufficient individual purchase histories, stores can predict upcoming grocery lists for the next time people visit the store using machine learning. This would solve the customer's issue of forgetting to buy certain products and provide extra convenience. In addition, there is a room for providing product recommendations based on what other people buy as well as suggesting cooking recipes accordingly. For example, imagine after a long day you decide to drive by the store because your food supplies are running low, but you don't remember what exactly you need to buy. Well, the store has you covered. Not only they will tell you which products you have to buy but also which additional products you should try because other people with similar preferences loved it. And when you feel adventurous with trying new recipes take a look on what you can make based on what ingredients you usually buy (TBD).

This idea, potentially, could offer additional convenience to customers. As for the retail chains, by providing aforementioned services, a flow of additional data from new loyalty card members will improve the cost savings, increase the revenue by individually suggesting new products to purchase and more importantly, when people forget to buy something, this is considered a forgone income for the store but with the system in place, chunk of that income can be 'reclaimed'.

Imagine you walk into a store and pull up the app powered by your favourite grocery chain. From there you can choose 2 options:

• predicted grocery list
• predicted list with some number of recommendations

Note: the entire app was deployed for demonstration purposes using StreamLit, however, the use is limited to entering a valid User_ID from the database. Therefore, the demonstration is limited to a GIF.

Proper order of Jupyter Notebooks:

1. Exploratory Data Analysis
2. Baseline Methods
3. Logistic Regressions
4. Recommender System

Initial data can be obtained here: Instacart

If you have any questions/suggestions or you are currently hiring, don't hesitate to contact via LinkedIn