BAC under scrutiny

Context

Over the years, the results of the Baccalaureate exam provide a detailed picture of the performance of high schools in Romania. By analyzing historical data over several years, we can discover trends, the evolution of academic performance and, sometimes, results that stand out from the pattern.

In this context, the purpose of this problem is to statistically analyze the performance of high schools at the Baccalaureate, to build predictive models and to identify atypical results using Machine Learning methods.

Data description

To solve the requirements of this problem you have access to a structured dataset in the form of two csv files: train.csv and test.csv. This dataset contains aggregated results of high schools for different subjects of the Baccalaureate exam, for the period 2014–2023.

Each row in the dataset (whether it's the training set or the test set) represents the results of a high school, for a subject, in a year.

Available columns:

• id – unique identifier of the row
• an – exam year
• liceu – name of the high school
• judet – county where the high school is located
• materie – Baccalaureate subject
• medie – average of grades obtained by the high school students for that subject
• procent_reusita – percentage of students who passed the exam
• numar_candidati – number of candidates
• preferinta_materie – percentage of students who chose that subject
• anomalie - can have the value 0 if it's not considered an abnormal situation or the value 1 if the situation is unusual.

Note: For each high school we also have some general statistics. For those, in the materie column we find the value GENERAL.

The train.csv file contains statistics from the period 2014-2022, and the test.csv file contains data for the year 2023, but the medie and anomalie columns are not available.

Requirements

Requirement 1 - Identifying the preferred subject (10 points)

Identify which was the subject with the highest preference in 2023 at COLEGIUL NATIONAL "UNIREA" FOCSANI, if we exclude the subject LIMBA ROMANA which was mandatory for all candidates.

Requirement 2 - Identifying the prosperous year for Computer Science (10 points)

Identify in which year from the period 2014-2022 the subject INFORMATICA MI C-C++ had the highest popularity at COLEGIUL NATIONAL "UNIREA" FOCSANI. We consider that popularity is given by the percentage from the preferinta_materie column (the higher this percentage, the more popular we consider the subject was).

Requirement 3 - Estimating high school performance (40 points)

Using the information available in the train.csv file, estimate the average grade obtained for each row in the test.csv file (year 2023).

For this requirement you can use any statistical analysis or modeling method based on historical data, taking into account:

• the analyzed subject,
• the high school's history,
• the evolution over time of results,
• the general context of performance at county or national level.

The goal is to obtain estimates as close as possible to the real values.

Requirement 4 - Identifying atypical results (40 points)

Analyze the historical data from the period 2014–2022 and decide, for each row in the test.csv file, whether the corresponding result for 2023 is:

• a normal one, or
• an atypical one, which deviates significantly from usual behavior.

In making the decision you can take into account:

• the history of the respective high school,
• variations from one year to another,
• differences between subjects,
• comparison with other similar high schools.

The final result for each row must be a binary label:

• 0 - normal result
• 1 - atypical result

Submission format

For automatic evaluation you must upload a file in csv format with the following structure:

• id – row identifier (corresponding to the one in test.csv)
  • for task 1, you will use the value 1 for id
  • for task 2, you will use the value 2 for id
• subtaskID – requirement identifier:
  • 1 for identifying the preferred subject (requirement 1)
  • 2 for identifying the prosperous year for Computer Science (requirement 2)
  • 3 for estimating the average (requirement 3)
  • 4 for identifying atypical results (requirement 4)
• answer – answer corresponding to the requirement:
  • for subtaskID = 1: the name of the subject exactly as it appears in the dataset
  • for subtaskID = 2: the year
  • for subtaskID = 3: a numerical value (average estimation)
  • for subtaskID = 4: 0 or 1

Evaluation

For requirements 1–2 evaluation is exact (by comparison).

For requirement 3 evaluation is done using Mean Absolute Error (MAE).

Rules:

• MAE < 0.3 → maximum score
• MAE > 1.0 → 0 points
• For values between these two thresholds, the score is awarded proportionally.

For requirement 4, the score is awarded according to the following rules:

• AUC ≥ 0.95 → 40 points
• AUC ≤ 0.70 → 0 points
• The interval between them: proportional score

⬇️ Download Starter Kit