/fire-prediction

Notebook for the Fire fighting using data on Zindi. Ranked number 5 on the public leaderboard and 8 on the private leaderboard. https://zindi.africa/hackathons/cmu-africa-fighting-fire-with-data

Primary LanguageJupyter Notebook

CMU-Africa: Fighting Fire with Data

Competition held on Oct. 31, 2020

Brief summary

The task consisted in predicting the percentage of area burnt for over specific periods of month using the dataset provided. To solve the regression problem, I essentially did some feature engineering, combining some features together and applying functions to others to draw more insights (exp, log, x^2, etc.). By proceeding like that, I was able to get a RMSE of 0.23 on the public leaderboard with the RidgeCV algorithm.

A series of boosting algorithms have then been tried and then stacked with the linear regression model to achieve the final position on the leaderboard.

A few more feature engineering would certainly have yielded positive results along with hyperparameter tuning that I could not finish on time due to time constraints.