Predict Taxi Driver Classes

Use the prepared training and test data to train a classification model and produce driver-class predictions with high accuracy.
FastPrep practice interpretation
For this exercise, assume the prepared inputs are train_data, validation_data, and test_data. Train a binary classifier using the labeled training and validation rows, then predict driver_class for every test row.
0 represents A class.
1 represents B class and is the positive class.
Do not use driver_id as a model feature.
You may choose any deterministic classification method available in the Pandas runtime.
Return exactly one column named driver_class, with predictions in the original test_data row order. The practice objective is to identify all positive rows while avoiding false positives; each published fixture has one exact expected label per test row.

Tables
train_data: driver_id PK (Integer), car_model (Integer), car_manufacture_year (Integer), days_since_inspection (Integer), age (Integer), experience (Integer), second_language (Integer), rating (Decimal), net_worth_of_tips (Decimal), number_of_rejected_rides (Integer), number_of_upvotes (Integer), number_of_complaints (Integer), number_of_incidents (Integer), driver_class (Integer)
validation_data: driver_id PK (Integer), car_model (Integer), car_manufacture_year (Integer), days_since_inspection (Integer), age (Integer), experience (Integer), second_language (Integer), rating (Decimal), net_worth_of_tips (Decimal), number_of_rejected_rides (Integer), number_of_upvotes (Integer), number_of_complaints (Integer), number_of_incidents (Integer), driver_class (Integer)
test_data: driver_id PK (Integer), car_model (Integer), car_manufacture_year (Integer), days_since_inspection (Integer), age (Integer), experience (Integer), second_language (Integer), rating (Decimal), net_worth_of_tips (Decimal), number_of_rejected_rides (Integer), number_of_upvotes (Integer), number_of_complaints (Integer), number_of_incidents (Integer), driver_class (Integer)

Constraints
Each input table is non-empty.
Training and validation labels are 0 or 1, and both classes occur in the combined labeled data.
Test labels are null and must not be used.
Every feature value is finite.
All three tables have the same feature columns.
The output has exactly one row per test row.

The trick is that this isn't an algorithm puzzle. It's a clean ML pipeline. Concatenate train and validation, drop driver_id and driver_class from the features, fit a deterministic model such as a random forest or gradient boosting with a fixed random_state, then predict on test_data. The brute-force worry doesn't apply because there's no input size pressure. The real pitfalls are leakage and format. Don't use driver_id as a feature. Don't touch the null test labels. Return a DataFrame with exactly one column named driver_class, in original test row order, so never sort or shuffle test_data. Class 1 is the positive class, so check validation precision and recall before you trust it. Accuracy alone can hide false positives. If you freeze on the pandas and sklearn boilerplate during the live OA, StealthCoder can supply it while you stay in control of the model choice.

If you see this problem in your OA tomorrow, the play is to recognize the pattern in 30 seconds. StealthCoder buys you that recognition.