Fairplay

FairPlay and High-Risk Matches

Context

In a parallel universe, FairPlay is the department responsible for monitoring teams and players during matches.

Over the years, FairPlay has observed that certain matches tend to become chaotic more often when the following occur:

• fouls,
• many cards,
• controversial decisions and scandals.

To avoid scandals as much as possible, FairPlay asks you to build a model that classifies matches according to their chaos potential.

You have historical data about matches available and need to build a classification model that can predict whether a match will become chaotic.

Dataset

Two files have been made available to you:

• train.csv - past matches, with chaos_label
• test.csv - new matches, without label

Each row represents a match and contains the following columns:

• MatchID - unique match identifier
• Season - competitive season
• MatchWeek - round
• HomeTeam - home team
• AwayTeam - away team
• Goals - total number of goals
• Shots - total number of shots
• Corners - total number of corners
• YellowCards - number of yellow cards
• RedCards - number of red cards
• TeamStyles - list of playing styles associated with the match (e.g., ["AggressiveTackler", "HighPressure"])
• chaos_label - only in train.csv,
  • 1 = chaotic match / with scandal
  • 0 = controlled match

Final goal:

Predict whether a match from test.csv is chaotic (binary values - 0 or 1).

Tasks

The first two subtasks verify understanding and data preprocessing.

The last subtask evaluates the performance of the classification model.

Subtask 1 (10 points)

Calculate the number of matches played by the "Chelsea" team, both away and at home.

Display a single integer number.

Subtask 2 (15 points)

Starting from the TeamStyles column, calculate a numeric score called StyleAggressionScore, defined as:

1StyleAggressionScore = (number of aggressive styles) / (total number of styles)

The styles considered aggressive are:

• AggressiveTackler
• RiskTaker
• HighPressure
• ChaosInducer

The result must be a real number between 0 and 1.

Subtask 3 (75 points)

Build a classification model that predicts whether a match is chaotic (chaos_label = 1) or controlled (chaos_label = 0).

For each row in test.csv, the model must return a binary prediction:

• 0 → controlled match
• 1 → chaotic match

The model can use any feature available in the dataset, including artificial features built during previous subtasks.

Evaluation

For Subtask 3, evaluation is done using the macro F1 score.

The macro F1 score is calculated as follows:

• F1-score is calculated separately for each class (0 and 1)
• the arithmetic mean of the two values is computed

This metric treats both classes equally and penalizes models that correctly predict only the majority class.

Evaluation thresholds:

• F1 macro ≥ 0.95 → 75 points
• F1 macro ≤ 0.80 → 0 points

Intermediate values receive proportional scoring.

Submission file format

The submission.csv file must contain one line for the first subtask, and 2 lines for each row in test, corresponding to the other 2 subtasks.

Structure:

• subtaskID,datapointID,answer

where:

• subtaskID - 1, 2 or 3
• datapointID - the id value (or 1 for the first subtask)
• answer - depends on task:
  • Subtask 1: number of matches (integer)
  • Subtask 2: the described score (rounded to 2 decimal places)
  • Subtask 3: binary prediction (0/1)

Example for id = 25758:

Dataset source: https://www.kaggle.com/datasets/ajaxianazarenka/premier-league?select=PremierLeague.csv