k-fold cross-validation in Machine Learning

K-fold cross-validation technique is widely used in Machine learning industry to evaluate the model performance.

How k-fold cross valiadtion works?

Step 1: Divide the data into k non-overlapping buckets randomly.

 

Step 2: The training and validation process takes 'k' iterations, with each iteration we use one of the 'k' buckets as the validation set and the remaining ('k-1') buckets as the training set.

 

Step 3: Calculate the performance metrics for each iteration.

 

Step 4: Once we have metrics for each iteration, we can use a variety of approaches like Mean accuracy, cross entropy, by visualizing the results on each iteration etc., to select the best model.

 

For example, if we have a dataset with 1000 data points and we use k-fold cross-validation with k=5, then the dataset will be split into 5 folds of 200 samples each. The model will be trained on 4 of the folds/buckets and evaluated on the remaining fold. This will be repeated 5 times, and the average of the 5 performance metrics will be reported.

 

In the iteration 1, we can use

1. fold1 data for the evaluation
2. fold2, fold3, fold4 and fold5 data for the training 

Is K-fold cross-validation works for both classification and regression?

Yes, k-fold cross validation works for both classification and regression problems.

Previous                                                    Next                                                    Home