Mirai9 | Devpost

React Home page
React page with Predictions from user input
ML dataset that was trained
precision recall curve, ROC curve, precision recall by threshold
confusion matrix
feature importance, showing which feature impacted model training
Endpoint with example for predicted value

🚀Inspiration

Stakes, friendly fights, and fiery debates have followed since the 2022 FIFA world cup's impending ascent and enthusiasm. Our team made the decision to apply machine learning principles to real data in order to assess and forecast which side would win these matches.

🤔What it does

Our front-end web app allows users to match-up 2 countries that are participating in the 2022 FIFA world cup. Then it will display the win percentage from our back-end API which is in constant communication with an AI. This AI was taught through Machine Learning with a large dataset of past soccer games and their results. Using extensive services like Google Cloud in order to implement Vertex AI, we are able to train this large dataset in order to obtain the results for a win, a draw, or a tie given the two team names. Using Google Clouds Vertex AI we were able to deploy endpoints in order to then work with the predicted data in our Flask app. This Flask app communicated with our React app in order to display the information according to user input, and team matchup combinations.

👨‍💻How we built it

We built the front end using React, and bootstrap. The front end communicates with an endpoint running Flask and Python. This is because we need to have an authenticated computer talk directly to the Google Cloud. This endpoint calls Google Cloud API, which predicts the odds of a team winning, given some information.

🚧Challenges we ran into

From comprehending how it operates to building our own machine learning environments and processing information through them, machine learning presented a new challenge for all of us. A challenge was figuring out how to apply Vertex AI's AutoML to our dataset and modify it to make it more compatible with the AI. Another challenge was our React app. Learning React and building our first web application was challenging, and in order to make our React website work best with backend computation, we had to cram a lot of tutorials and documentation into it.

🎯Accomplishments that we're proud of

making an AI that responds to the right request and outputs the right data We entered this data into Google Clouds Vertex AI, where we trained and successfully predicted soccer match results given the names of two teams utilising the effective and user-friendly AutoML technology.

💡What we learned

Working with Google Clouds Platform our team really extended our capabilities across the wide array of features that it included, all the way from creating VM instances to host python flask apps, to utilizing the Vertex Ai’s AutoML to train our large dataset, and effectively target specific columns to produce the data that we needed. In order to understand and effectively communicate the data in a visual manner our team had to learn react, and understand HTTP requests, this process involved long hours of research and team efforts.

📈What's next for Mirai9

Implementing our machine learning model to more variety of sports, esports, and even hackathon statistics (using data on ideas) to predict which projects have higher chances of winning based on prior hackathon data. Mirai 9 works to achieve more accurate predictions in the near future with greater success than as of current.