Forest Cover Classification - ML Hack'19
Successful forest cover type classification has a lot of potential for positive change, particularly in areas like environmental conservation, flora and fauna research, and geological studies. Automated data-driven low-cost solutions in ecology have a huge scope to reduce the domain expertise required by Forest reserve officials. An end-to-end ML based system can help get a fair estimate of the type of damage caused by wildfires in a shorter span of time and help the officials in various ways. While collecting ecological data, it's quite possible that errors can creep in and thus there's a need for robust solutions for classifications.
Here, you'll need to classify whether an annotated tree is a Western Yellow Pine or not.
https://www.kaggle.com/c/cte-ml-hack-2019
- Designed an end-to-end DL based system for classifying forest covers.
- The model was a 8-layer feedforward neural network, regularized using dropout and trained for 150 epochs. It achieved a accuracy of 0.96380 on the private leaderboard (test set), that saw me finish runner up.