Flop3_datathon_2025_schneider_project

KNN feature importance

Inspiration

We are first year first semester students passionate about learning new skills within the engineering industry, this datathon seemed like the best fit to challenge ourselves to the unknown.

What it does

Our project consists of 3 folders: the first one contains a code to test 10 ML models and print out the metrics of each model then we chose 3 of the best performing models(RandomForest, XGBoost, KNN), and compare the variance, for our main code which is in another folder(final_demo.py), the third folder stores the images of the graphs needed for the data analisis. We wanna note that the full run throught on the final_demo.py may need 50-70min for it to output

How we built it

We used a variety of libraries with different functions to help us navigate throught the the ~36k data samples, the main code final_demo.py trains all 3 models and finds the different insights on each of the model KNN, XGBoost, RandomForest: Feature importance and Partial Dependence Plot(PDP). XGBoost: shap value. XGBoost, Random Forest: 5 random lime_exp.

Challenges we ran into

Our main challenge was that none of us had previous formation on the needed skills for the datathon, we had to improvise, learn and use every tool we could including: adjunted book on the instruccions, AI tools and many online resources like youtube, geeks for geeks, etc.

Accomplishments that we're proud of

In a marathon mindset we worked consistently and intensely in the given 26h of time, we had fixed many errors and archived something we thought was far away from us, we enjoyed the datathon, met a lot of people with similar goals and formation, the struggle payed of as a compressed zip file.

What we learned

We went from just knowing for-loops to understanding basic machine learning models, using graphing and analysis libraries and being able to work in small groups, dividing the tasks, switching places to work and collaborate as programmers.

What's next for Flop3_datathon_2025_schneider_project

We would have loved to use more insighs and had more time to play with the data and the parameters to analyse the different outcomes, we plan to participate the next year, hopefully we will be better formed for the new challenges we will face.