Train-test split for time series in ML Python - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the main difference between train-test split for time series data and for regular data?

In time series, the data is ordered by time, so the train-test split must keep this order to avoid using future data to predict the past. Regular data can be shuffled before splitting.

beginner

Why should you never randomly shuffle time series data before splitting into train and test sets?

Random shuffling breaks the time order and can cause the model to learn from future information, which is unrealistic and leads to over-optimistic results.

beginner

What is a common method to split time series data into train and test sets?

Use the earliest part of the data for training and the later part for testing, preserving the time order.

intermediate

How does the size of the test set affect time series model evaluation?

A larger test set gives a better estimate of future performance but reduces training data size. A balance is needed to train well and evaluate reliably.

intermediate

What is the purpose of using a rolling or expanding window approach in time series train-test splitting?

These approaches simulate real forecasting by repeatedly training on past data and testing on the next time step, helping to evaluate model stability over time.

Why can't you randomly shuffle time series data before splitting into train and test sets?

A It breaks the time order and leaks future information into training

B It makes the dataset too small

C It increases the training time

D It improves model accuracy

Random shuffling breaks the time order and causes the model to learn from future data, which is not realistic.

What is the typical way to split time series data for training and testing?

A Randomly split 50% train and 50% test

B Use the earliest data for training and the latest data for testing

C Shuffle data then split

D Use only the last data points for training

Time series data must keep time order, so training uses earlier data and testing uses later data.

What does a rolling window approach do in time series model evaluation?

A Ignores time order

B Randomly selects data points for training

C Uses only the first half of data for training

D Trains and tests repeatedly on moving time windows

Rolling window trains on a fixed-size past window and tests on the next time step repeatedly.

What is a risk of using too small a training set in time series?

A Model will overfit perfectly

B Test set will be too large

C Model may not learn enough patterns

D Data will be shuffled

A small training set may not provide enough information for the model to learn well.

Which of these is NOT a valid reason to keep time order in train-test split for time series?

A To increase randomness in training data

B To simulate real forecasting scenarios

C To avoid data leakage from future to past

D To evaluate model on unseen future data

Increasing randomness by shuffling is not valid for time series splitting.

Explain why preserving time order is important when splitting time series data into train and test sets.

Think about how time flows and why using future data to predict past is a problem.

Describe how a rolling window approach works for training and testing time series models.

Imagine sliding a small window over your data to train and test repeatedly.