HackUTD2022 EOG challenge

DevPost = https://devpost.com/submit-to/16723-hackutd-ix/manage/submissions/370578-d-day/project_details/edit

Given a wide variety of data, with an open ended prompt, we were excited to use classic unsupervised and supervised learning techniques to classify techniques.

D-Day is a website that takes in data about asteroid drills sent in by the user and uses ML techniques to present accurate and useful analytics. The code promotes and very modular design for reproducibility and easy access. Our website design illustrates flawless user interface for the most efficient, streamlined process possible.

We utilized primarily Python, Numpy, Pandas, and Scikit-learn for the true processes behind our project. We used a variety of mathematical and statistical techniques in order to normalize and transform the given data in order to achieve the desired results. We made completely modular functions in order to allow the user to select any metric, any feature, any combination, and any possible set of data and return an accurate result. We were able to rank the drill bits by cost and performance. We were able to create a viable feature space to compare different combinations of features. We were also able to get the correlation between features and how they affected each other. In addition, we were able to create an easy to use UI/UX specifically designed to make the user's experience quick and efficient

One problem we faced was the numerous occurrences of bad data which we needed to clean. Most of our time went into creating features and doing routine data processing in order to create a strong foundation for our model. In addition, we faced the challenge of selecting a model in order to best fit the problem. While we could have used some sort of CNN and deep learning for the base of the project, we decided it would be much more efficient and cost-smart to utilize a simpler model such as a K-nearest neighbors algorithm.

We were able to answer all of the given questions given as part of the project prompt. In addition, we were able to successfully devise a way to monitor and check live user data through the functionality of data submission on the website. We were able to normalize and process the results in order to make an readable output for any user. We were able to determine that the best drilled Asteroid was Asteroid #3 and that the most efficient drill-bit per depth/time was the "Apollo Drill-bit."

We learned even further about the importance of data preprocessing and cleaning. Furthermore, we learned more about the variety of solutions that occur per any problem. I believe we advanced our ML and AI skills by a significant degree while also maintaining respect for UI/UX and frontend.

The next step for our project in fully fleshing out and connecting our website. Moreover, we thought that an efficient way to generate the closest permutation of specific labels would be through reinforcement learning, a system based on weights and rewards. While we were unable to fully realize this approach, we believe we have enough understanding of this technique in order to more accurately solve this problem.