K-Fold Cross-validation Complete Guide — How To Tutorial In Machine Learning & Deep Learning

by | Jul 29, 2023 | Data Science, Machine Learning

What is k-fold cross-validation?

K-fold cross-validation is a popular technique used to evaluate the performance of machine learning models. It is advantageous when you have limited data and want to maximize it while estimating how well your model will generalize to new, unseen data.

The basic idea behind k-fold cross-validation is to split the dataset into k subsets of approximately equal size, or “folds.” The model is trained and evaluated k times, using a different fold as the validation set and the remaining k-1 folds as the training set. The final evaluation metric is usually the average of the evaluation results from all k iterations.

Example of a k fold cross-validation split with k=4

Example of a k fold cross-validation split with k=4

Step-by-step explanation of the k-fold cross-validation process

  1. Data Preparation: Start with a dataset you want to use for training and evaluation.
  2. Shuffling (Optional): Shuffling the data randomly before splitting it into folds is an excellent practice. This helps reduce potential biases due to any inherent order in the data.
  3. Partitioning: Divide the data into k subsets of approximately equal size. These subsets are called “folds.”
  4. Model Training and Evaluation: For each fold:
    a. Use k-1 folds as the training data to train the machine learning model.
    b. Use the remaining 1 fold as the validation data to evaluate the model’s performance.
    c. Record the evaluation metric (e.g., accuracy, precision, recall, F1-score).
  5. Performance Metric Calculation: After performing k iterations, calculate the average of the evaluation metrics obtained during each iteration. This estimates how well the model is expected to act on unseen data.
  6. Model Selection: Depending on your objective, you may choose the best-performing model from the k iterations, or you can use this process to fine-tune hyperparameters, architecture, or other aspects of your model.

Advantages of k-fold cross-validation

  • It provides a more robust estimate of the model’s performance since it uses multiple data splits.
  • Makes efficient use of data, especially when the dataset is small.

Common choices for k in k-fold cross-validation are 5 and 10, but you can choose other values based on your specific dataset and requirements. However, keep in mind that as k increases, the computational cost also increases. In extreme cases, when k is equal to the number of samples in the dataset (k=N, known as leave-one-out cross-validation), each sample is used as a separate validation set, but this can be computationally expensive for large datasets.

Remember that the primary goal of cross-validation is to estimate how well your model generalizes to new, unseen data, and it can help you identify potential issues like overfitting or underfitting.

How does cross-validation help avoid overfitting and obtain a more reliable model performance estimate?

Cross-validation helps in avoiding overfitting and obtaining a more reliable model performance estimate in the following ways:

  1. Reduced Overfitting: Overfitting occurs when a model performs well on the training data but poorly on unseen data. With traditional train-test splitting, there’s a risk that the model’s performance estimate is overly optimistic because it has seen the test data during training. Cross-validation mitigates this by using multiple train-test splits and averaging the results, reducing the impact of any specific data split. By repeatedly training on different subsets of the data, cross-validation provides a more balanced and generalizable assessment of the model’s performance.
  2. Better Utilization of Data: In traditional train-test splitting, a portion of the data is reserved for testing, which reduces the amount of data available for training. Cross-validation addresses this by using the entire dataset for both training and testing in different folds. This leads to better utilization of the data and more reliable performance estimates, especially when data is limited.
  3. Hyperparameter Tuning: Cross-validation is commonly used for tuning, where different hyperparameter values are evaluated and compared. By using cross-validation for this process, you avoid selecting hyperparameters based on their performance on a single train-test split, which may not represent the model’s generalization capability.
  4. Model Selection: When comparing multiple models or algorithms, cross-validation provides a fair and consistent evaluation, ensuring all models are assessed on the same data splits. This helps select the best-performing model based on its average performance across multiple folds.
  5. Robustness to Data Variability: Datasets may have inherent variability due to the randomness of data collection or external factors. By using different data splits in cross-validation, the model’s performance estimate becomes more robust and less sensitive to variations in the dataset.
  6. Insight into Model Variance: By analyzing the variance in performance across different folds, you can gain insights into the model’s sensitivity to different training data. If the variance is high, it may indicate that the model is not stable and might benefit from more data or regularization techniques.

Overall, cross-validation provides a more comprehensive evaluation of the model’s performance by repeatedly assessing its generalization capabilities on multiple subsets of the data. This process helps in selecting the best model, avoiding overfitting, and gaining more confidence in the model’s ability to perform well on unseen data.

What are the other variations of cross-validation techniques besides k-fold cross-validation in machine learning?

Cross-validation in machine learning is a model evaluation technique to assess how well a machine learning algorithm will generalize to new, unseen data. The goal is to estimate the model’s performance on data not seen during the training phase. Cross-validation is especially useful when you have limited data or want to obtain a more reliable evaluation of your model’s performance.

The most common type of cross-validation is k-fold cross-validation. However, there are other variations of cross-validation techniques, such as stratified k-fold, leave-one-out, and leave-p-out cross-validation. Let’s briefly discuss some of these techniques:

  1. K-Fold Cross-Validation: The dataset is divided into k subsets (folds), and the model is trained and evaluated k times. Each time, one fold is used as the validation set, and the remaining k-1 folds are used as the training set.
  2. Stratified K-Fold Cross-Validation: This variation is commonly used for classification tasks where the class distribution is not uniform. A stratified k-fold ensures that the class distribution is preserved in each fold, reducing potential biases in the evaluation.
  3. Leave-One-Out Cross-Validation (LOOCV): In this approach, k is set to the number of samples in the dataset (N). Each sample is used as a separate validation set, and the model is trained on N-1 samples. LOOCV is computationally expensive but provides an unbiased model performance estimate.
  4. Leave-P-Out Cross-Validation: Similar to LOOCV, but instead of leaving one sample out, p samples are left out for validation, and the model is trained on the remaining data.
  5. Time Series Cross-Validation: This technique is used when dealing with time series data, where the order of the data points matters. Time series cross-validation ensures that the model is trained on past data and evaluated on future data, simulating real-world scenarios. We will discuss this technique in more detail further on.

Cross-validation is crucial for avoiding overfitting and obtaining a more reliable model performance estimate. After performing cross-validation, you can examine the average performance metric (e.g., accuracy, mean squared error, etc.) to assess how well your model will likely perform on new, unseen data. It also helps to tune hyperparameters and identify potential issues with the model’s generalization capability.

How can one choose the right k for cross-validation?

Choosing the correct value for k in k-fold cross-validation can impact the performance estimation of your model and the overall efficiency of the evaluation process. The selection of the appropriate k value depends on various factors, including the size of your dataset, the distribution of the data, and the computational resources available. Here are some guidelines to help you decide the correct k value:

  1. Smaller Datasets: For small datasets, using a larger k value (e.g., k=5 or k=10) is preferred. This allows more data for training and validation in each fold, reducing the variance in performance estimates.
  2. Larger Datasets: For larger datasets, you can use a smaller k value (e.g., k=3) to speed up the cross-validation process while obtaining a reasonable estimate of model performance.
  3. Imbalanced Datasets: If your dataset is imbalanced, meaning the class distribution is not uniform, consider using stratified k-fold cross-validation (or variants like StratifiedKFold) to ensure that each fold preserves the class distribution. This helps to prevent biased evaluation and provides more reliable performance estimates, especially in classification tasks.
  4. Time Series Data: For time series data, where the order of observations matters, consider using time series-specific cross-validation techniques like TimeSeriesSplit, which ensure that the model is evaluated on future data points after being trained on past data points.
  5. Computational Resources: Larger k values lead to more model training iterations, making cross-validation more computationally expensive. If you have limited computational resources, it may be more practical to use smaller k values.
  6. Trade-off Between Variance and Bias: Smaller k values lead to a higher variance in performance estimates, while larger k values may introduce a higher bias. Sticking a balance is essential based on the specific problem and data.
  7. Nested Cross-Validation: If you are using cross-validation for hyperparameter tuning (e.g., with GridSearchCV), you can perform an outer cross-validation for model evaluation (e.g., k=5) and an inner cross-validation for hyperparameter tuning (e.g., k=3). This technique, known as nested cross-validation, can provide more robust estimates when dealing with limited data.

In practice, it’s a good idea to experiment with different k values and compare the results. You can perform a grid search over k values and assess how they impact the model’s performance. Ultimately, the choice of k will depend on the specific characteristics of your dataset and the objectives of your analysis.

How can you carry out cross-validation for time series?

Time series cross-validation is a specific type of cross-validation used for evaluating machine learning models on time series data. In time series data, the order of observations matters as each data point is recorded at a specific time. Therefore, the typical random splitting used in traditional cross-validation may not be suitable for time series datasets, as it can introduce data leakage and unrealistic evaluation scenarios.

The main idea behind time series cross-validation is to mimic the real-world scenario where the model is trained on historical data and tested on future data without using future information during training. This approach provides a more realistic evaluation of the model’s performance and ability to predict unseen future data points.

There are two standard methods of time series cross-validation:

  1. Train-Test Split with a Time-based Cut-off: In this approach, you split the time series data into two parts: the training set, which contains data up to a specific time point, and the test set, which includes data from a later time point. The model is trained on the training set and evaluated on the test set. The time-based cut-off ensures that data from the test set comes after the data used in the training set.
  2. Rolling Window Cross-Validation: This method uses a fixed-sized rolling window to create multiple train-test splits. The window slides through the time series, and at each step, you use a portion of the data as the training set and the subsequent data as the test set. This process continues until the entire time series is covered. It’s like performing multiple train-test splits with a sliding window.

Both methods ensure that the model is not exposed to future information during training, and the evaluation reflects the model’s ability to make predictions on future data points.

It’s important to note that when using time series cross-validation, the order of the data should be preserved, and you should be cautious not to introduce any data leakage. Data leakage can occur when features or information from the future are used during training, leading to overly optimistic performance estimates.

Time series cross-validation is valuable for selecting appropriate hyperparameters, assessing model performance, and gaining insights into how well a model generalizes to future observations. However, it is computationally more demanding than traditional cross-validation due to the sequential nature of the data, as the model needs to be retrained for each validation fold.

Tutorial: K fold cross-validation with sklearn in Python

In Python, you can perform cross-validation using various libraries, but one of the most commonly used libraries for this purpose is scikit-learn (sklearn). Scikit-learn provides a straightforward interface for implementing different cross-validation techniques. Below, we will show you how to perform k-fold cross-validation using scikit-learn:

import numpy as np
from sklearn.model_selection import KFold
from sklearn.model_selection import cross_val_score
from sklearn import YourMachineLearningAlgorithm  # Replace with the specific algorithm you want to use

# Assuming you have your data in X (feature matrix) and y (target vector)
# Initialize your machine learning algorithm (e.g., a classifier or regressor)
model = YourMachineLearningAlgorithm()

# Set the number of folds for cross-validation
num_folds = 5

# Create a k-fold cross-validation object
kfold = KFold(n_splits=num_folds, shuffle=True, random_state=42)

# Perform cross-validation and obtain the scores (evaluation metrics)
scores = cross_val_score(model, X, y, cv=kfold)

# Print the results of each fold and the average performance
for fold_idx, score in enumerate(scores):
    print(f"Fold {fold_idx + 1}: Score = {score:.4f}")
print(f"Average Score: {np.mean(scores):.4f}")

In the above code, we use scikit-learn’s KFold class to create a k-fold cross-validation object with the desired number of folds (num_folds). The shuffle=True parameter ensures that the data is randomly shuffled before splitting into folds, reducing potential biases. The random_state parameter provides reproducibility by setting a random seed.

We then use cross_val_score to perform the cross-validation. This function takes the machine learning model (model), feature matrix (X), target vector (y), and the cross-validation object (cv=kfold). It returns an array of scores, where each score represents the evaluation metric (e.g., accuracy, mean squared error, etc.) obtained for each fold.

Remember to replace YourMachineLearningAlgorithm with the specific machine learning algorithm you want to use, such as DecisionTreeClassifierRandomForestRegressorLogisticRegression, etc., depending on the type of problem you are working on (classification or regression).

Scikit-learn also provides other cross-validation techniques, such as StratifiedKFoldTimeSeriesSplit, and LeaveOneOut, which you can use depending on your dataset and specific requirements.

K fold cross-validation with deep learning

Cross-validation can also be used with deep learning models to evaluate their performance and tune hyperparameters effectively. The process is similar to traditional machine learning models, but there are a few essential considerations specific to deep learning:

  1. Data Preprocessing: Deep learning models often require more extensive data preprocessing, such as normalization, data augmentation, or feature scaling. Ensure to consistently apply the same preprocessing steps across all folds during cross-validation.
  2. Time and Resource Constraints: Deep learning models can be computationally expensive and time-consuming to train, especially on large datasets. When using k-fold cross-validation, consider the available computational resources and choose an appropriate value of k that balances computation time and model evaluation accuracy.
  3. Data Leakage: When using data transformations or augmentations, be cautious about data leakage. Ensure that any data augmentation or modifications are applied only to the training set during each fold and not to the validation or test sets to avoid introducing data leakage.
  4. Architecture and Hyperparameter Tuning: Cross-validation can be used for tuning hyperparameters of deep learning models, such as learning rate, number of layers, batch size, and regularization parameters. Consider using nested cross-validation to perform hyperparameter tuning alongside model evaluation.

Cross-validation with Keras

Here’s an example of performing k-fold cross-validation with a deep learning model using the Keras library, which is a popular deep learning framework in Python:

import numpy as np
from sklearn.model_selection import KFold
from keras.models import Sequential
from keras.layers import Dense
from keras.wrappers.scikit_learn import KerasClassifier
from sklearn.model_selection import cross_val_score

# Assuming you have your data in X (feature matrix) and y (target vector)

# Define the function to create your deep learning model
def create_model():
    model = Sequential()
    model.add(Dense(64, activation='relu', input_dim=X.shape[1]))  # Adjust the input_dim based on your feature dimensions
    model.add(Dense(32, activation='relu'))
    model.add(Dense(1, activation='sigmoid'))  # For binary classification, use 'sigmoid'; for multiclass, use 'softmax'
    model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
    return model

# Create a KerasClassifier using the model creation function
model = KerasClassifier(build_fn=create_model, epochs=10, batch_size=32)

# Set the number of folds for cross-validation
num_folds = 5

# Create a k-fold cross-validation object
kfold = KFold(n_splits=num_folds, shuffle=True, random_state=42)

# Perform cross-validation and obtain the scores (evaluation metrics)
results = cross_val_score(model, X, y, cv=kfold)

# Print the results of each fold and the average performance
for fold_idx, score in enumerate(results):
    print(f"Fold {fold_idx + 1}: Score = {score:.4f}")
print(f"Average Score: {np.mean(results):.4f}")

In this example, we use KerasClassifier to wrap the deep learning model. The create_model function defines the architecture of the deep learning model using the Keras Sequential API. Adjust the model architecture and hyperparameters according to your specific problem.

Remember to preprocess the data appropriately for deep learning, such as scaling or normalization, and choose the appropriate loss and activation functions for your specific problem (e.g., binary classification, multiclass classification, regression, etc.).


Cross-validation is a fundamental technique in machine learning and deep learning used to assess the performance of models on unseen data. It helps evaluate the model’s generalization ability and provides more reliable performance estimates than a single train-test split.

The k-fold cross-validation method is widely used, where the dataset is divided into k subsets (folds), and the model is trained and evaluated k times. Each fold is the validation set once, and the remaining folds are used for training. The final evaluation metric usually averages the results obtained in all k iterations.

When using cross-validation with deep learning models, it is essential to be mindful of data preprocessing, potential data leakage, computational resources, and hyperparameter tuning. Deep learning models can be computationally expensive, so choosing an appropriate k value is crucial to balance accuracy and computational efficiency. Data preprocessing and augmentation should be applied consistently across all folds to prevent data leakage.

Overall, cross-validation is a valuable tool to assess the generalization performance of your models and make informed decisions regarding hyperparameter tuning and model selection. By using cross-validation, you can gain more confidence in the robustness and reliability of your machine learning and deep learning models.

About the Author

Neri Van Otten

Neri Van Otten

Neri Van Otten is the founder of Spot Intelligence, a machine learning engineer with over 12 years of experience specialising in Natural Language Processing (NLP) and deep learning innovation. Dedicated to making your projects succeed.

Recent Articles

One class SVM anomaly detection plot

How To Implement Anomaly Detection With One-Class SVM In Python

What is One-Class SVM? One-class SVM (Support Vector Machine) is a specialised form of the standard SVM tailored for unsupervised learning tasks, particularly anomaly...

decision tree example of weather to play tennis

Decision Trees In ML Complete Guide [How To Tutorial, Examples, 5 Types & Alternatives]

What are Decision Trees? Decision trees are versatile and intuitive machine learning models for classification and regression tasks. It represents decisions and their...

graphical representation of an isolation forest

Isolation Forest For Anomaly Detection Made Easy & How To Tutorial

What is an Isolation Forest? Isolation Forest, often abbreviated as iForest, is a powerful and efficient algorithm designed explicitly for anomaly detection. Introduced...

Illustration of batch gradient descent

Batch Gradient Descent In Machine Learning Made Simple & How To Tutorial In Python

What is Batch Gradient Descent? Batch gradient descent is a fundamental optimization algorithm in machine learning and numerical optimisation tasks. It is a variation...

Techniques for bias detection in machine learning

Bias Mitigation in Machine Learning [Practical How-To Guide & 12 Strategies]

In machine learning (ML), bias is not just a technical concern—it's a pressing ethical issue with profound implications. As AI systems become increasingly integrated...

text similarity python

Full-Text Search Explained, How To Implement & 6 Powerful Tools

What is Full-Text Search? Full-text search is a technique for efficiently and accurately retrieving textual data from large datasets. Unlike traditional search methods...

the hyperplane in a support vector regression (SVR)

Support Vector Regression (SVR) Simplified & How To Tutorial In Python

What is Support Vector Regression (SVR)? Support Vector Regression (SVR) is a machine learning technique for regression tasks. It extends the principles of Support...

Support vector Machines (SVM) work with decision boundaries

Support Vector Machines (SVM) In Machine Learning Made Simple & How To Tutorial

What are Support Vector Machines? Machine learning algorithms transform raw data into actionable insights. Among these algorithms, Support Vector Machines (SVMs) stand...

underfitting vs overfitting vs optimised fit

Weight Decay In Machine Learning And Deep Learning Explained & How To Tutorial

What is Weight Decay in Machine Learning? Weight decay is a pivotal technique in machine learning, serving as a cornerstone for model regularisation. As algorithms...


Submit a Comment

Your email address will not be published. Required fields are marked *

nlp trends

2024 NLP Expert Trend Predictions

Get a FREE PDF with expert predictions for 2024. How will natural language processing (NLP) impact businesses? What can we expect from the state-of-the-art models?

Find out this and more by subscribing* to our NLP newsletter.

You have Successfully Subscribed!