Develop an algorithm for classifying forest fires in the points of temperature anomalies
Competition of the classification algorithms of wild fires according to the data of temperature anomalies from satellites.
It is necessary to classify the type of a wild fire (according to the Ministry of Emergency Situations classification) based on the information about the point of temperature anomaly. The solution should be implemented in the form of a program, which takes a CSV table with points as input (latitude and longitude coordinates, and the date when the point was received). It is necessary to make a table with probabilities for each of 11 classes (columns fire_1_prob, fire_11_prob) as output.
The solution represents an archive with a code, which is launched in the Docker container environment. The solutions validation occurs at new points coming from an automatized system. The quality of the solution is evaluated by a set of points for a certain predetermined period of time. Quality metric — Micro-averaged Multi-Class ROC-AUC counted for all types of wild fires (11 classes).
Total prize pool is 700 000 RUB. The winning team will receive 300,000 RUB, the second place team will receive 200,000 RUB, the team placing the third will receive 100,000 RUB. In addition, the best public decision published on github will receive an additional jury prize of 100,000 RUB.