Determine a player's experience in the Dota 2 game based on information about his behavior during the match.
The game Dota 2 involves two sides: teams of the forces of Light (The Radiant) and the forces of Darkness (The Dire). Each team consists of 5 players - characters with their own unique abilities.
The goal of the game is to destroy the throne located in the enemy base.
Your task is to build an algorithm that can determine the player’s experience based on his telemetry - a layer of data that was prepared specifically for the in-person final. This is information about the player’s behavior during the match: detailed mouse movements, clicks, interaction with other players, time series.
In this task, it is necessary to estimate the probability of classes (0 - beginner, 1 - experienced), rather than “guessing” the correct answer — the ROC AUC quality metric.
For analysis, you can use advanced feature engineering techniques, neural networks and deep learning.
Each data example provided describes the characteristics of the match and the statistics of one of the players at the end of the match. All examples have a unique id.
Two data sets were prepared for the competition:
- Train, in which the player's experience medal type is known for each example.
- Test, where the participants do not know the player’s experience - it needs to be calculated.
Your task is to build an algorithm that estimates the probability of the skilled class. In the online stage task, the player's experience was determined by indicating the answer as 0 or 1, and the quality of predictions was measured using the Accuracy metric.
The task now is to estimate the chance (probability in the range [0, 1]) that the player is skilled. In practice, this means that the answer now indicates not just 1 if your algorithm predicted that the player is experienced, but some number close to 1, for example: 0.8742..
The constructed algorithm had to be run on test data, the result saved in a.csv-file and sent as your solution.
For each example from the test set, it was necessary to predict the probability of the skilled class. . It was necessary to send a CSV table with class 1 probability predictions to the system, which contained two columns: id - the player’s identifier, skilled_prob - the probability that he is experienced.
