Our inspiration came from us all being college students and not budgeting well or not having the money. It also came from the 84.51 Challenge with Kroger's data.
What it does
This application shows the customer their weekly spending at Kroger, what commodities they spend the most in. Users will have the ability to easily toggle through the weeks! Their information will then be mapped to both a bar & line chart. This will help users get a better budget break down thanks to their customer information provided by 8451 & Kroger! In addition to that, our application also incorporates Machine Learning to predict and estimate cost on how much you will spend given a certain number of commodities. More information on this is in the machine learning section. Our app brings a lot of value to those on a budget, looking to spend smarter!
How we built it
We built this application using React.js, Machine-Learning, Google-Cloud, Firebase, and more. We divide the work into two main parts. We had two people on front-end and two people on the machine learning aspect. The people on the machine-learning portion were responsible for deciding what data was real and usable for what we wanted to accomplish. The people on the front-end development portion were in charge of designing and creating the Web Application you see in the screenshots. We all collaborated and helped each other throughout the project.
More on Machine Learning
Our Machine Learning algorithm was generated & trained using Google Cloud Platform's AI Service Tables. The dataset we used was a derivative of the complete journey sample which we downloaded from 8451's area 51 website. The dataset consists of all the commodities that a single household would purchase in the given two-year span. This would total to around 400 rows of data (400 unique households). I'll make sure to put up the dataset. Now! As far as the Machine learning algorithm goes, the GCP platform trained a linearly regressive model. Some of the statistics go as follows:
- R^2 = 0.844
- MAE = 1,4720.91
- RMSE = 2,062.377
- MAPE = 35.54%
Googles Services also gave the importance of commodity features. The top 5 goes as follows:
Once our model was trained, totaling to a size of around 340MB & 1.5 Hrs of training time, we deploy the model. This entails generating a REST API to query the model which will reside in GCP's cloud services.
Challenges we ran into
The challenges we ran into were our original idea having to be scrapped and familiarizing everyone with all portions of the project. Our original idea was going to be a budgeting and nutritional app based off of Kroger Customers' spending. Sadly, the UPC Codes provided were not valid and made us have to rethink our idea.
Accomplishments that we are proud of
We were proud of that fact that we were able to learn more about google cloud and apply it to our application. It was amazing to see how courses at our university benefited us or allowed us to go above our basic knowledge, strive for higher goals, and achieve them. One person on our team was really excited to learn more about excel and how much you can really do with it. We are really proud of how our front-end turned out! It looked much better than we expected since the two people working on it have not really worked on React.js, if at all. Lastly, we were really proud of how much we learned about React.js and how we were able to implement it to do the functionality to make this application the best it can be in 24 hours.
What we learned
We all learned more about Machine-Learning and React as a whole. We learned how to be more efficient to be able to complete the project on time and accomplish all of our goals.
What's next for Super-Marka-Metrics
The goal is to move it into the nutritional market, using UPC codes, that helps users budget and stay healthy throughout their Kroger Journey!