Train-test split for time series in ML Python - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why is random train-test splitting not suitable for time series data?

In time series data, why should we avoid randomly splitting data into training and testing sets?

A Because random splitting breaks the time order, causing data leakage from future to past.

B Because random splitting reduces the size of the training set too much.

C Because random splitting always causes the model to overfit.

D Because random splitting makes the test set too small to evaluate.

💡 Hint

Think about how time flows and what it means to predict future values.

Random splitting mixes past and future data, letting the model see future information during training, which is unrealistic and causes data leakage.

❓ Predict Output

intermediate

Output of a time series train-test split code

What is the length of the training and testing sets after this split?

ML Python

import numpy as np
from sklearn.model_selection import train_test_split

data = np.arange(10)
train, test = train_test_split(data, test_size=0.3, random_state=42)
print(len(train), len(test))

A 3 7

B 7 3

C 10 0

D 6 4

💡 Hint

Check how train_test_split divides data by default.

train_test_split splits data randomly; with test_size=0.3 on 10 items, training has 7 and testing has 3 items.

❓ Model Choice

advanced

Best method to split time series data for forecasting

Which method correctly splits time series data to avoid data leakage and respect temporal order?

A Randomly shuffle data then split into train and test sets.

B Split data by selecting every other data point for testing.

C Use k-fold cross-validation with random folds.

D Split data by taking the first 80% as training and last 20% as testing.

💡 Hint

Think about keeping the time order intact.

Using the first 80% as training and last 20% as testing preserves time order and prevents future data leakage.

❓ Hyperparameter

advanced

Choosing test size for time series split

What is a key consideration when choosing the test size for a time series train-test split?

A Test size should be as small as possible to maximize training data.

B Test size should be random to avoid bias.

C Test size should cover a full seasonal cycle if seasonality exists.

D Test size should always be 50% for balance.

💡 Hint

Think about repeating patterns in time series.

Choosing a test size that covers a full seasonal cycle ensures the model is tested on all seasonal patterns.

🔧 Debug

expert

Identify the error in this time series split code

What error will this code raise when splitting time series data?

ML Python

import numpy as np
from sklearn.model_selection import train_test_split

data = np.arange(10)
train, test = train_test_split(data, test_size=0.3, shuffle=False)
print(train)
print(test)

A No error; train contains first 7 elements, test last 3 elements.

B TypeError because test_size must be an integer when shuffle=False.

C ValueError because shuffle=False is not allowed with train_test_split.

D IndexError because data is too small for test_size=0.3.

💡 Hint

Check sklearn docs for shuffle parameter behavior.

train_test_split allows shuffle=False; it splits data sequentially without error.