If you like what we're working on, please  Star Us on GitHub. This enables us to continue to give back to the community.

Model Selection


Machine Learning is, at its most basic level, the combination of statistics and computing. Machine learning is based on the notion of algorithms or models, which are statistical guesses taken to the next level.

However, depending on the data distribution, each proposed model has many drawbacks. Because they are only estimates, none of them can be completely correct. The terms “bias” and “variance” are commonly used to describe these constraints.

A model with a large bias will simplify too much by ignoring the training points.

By not generalizing for test points it hasn’t seen before, a model with high variance will limit itself to the training data.

When the limits are slight, for example when deciding between a random forest technique and a gradient boosting approach, or between two variants of the same decision tree algorithm, the problem emerges. Both will have a lot of variances and little bias.

Different types of the Model selection

After the different models have been reviewed based on the relevant criteria, model selection is a strategy for picking the best model.

Methods of resampling

Resampling methods are basic strategies for rearranging data samples to see if the model works well on data samples that haven’t been trained on. To put it another way, resampling allows us to see if the model will generalize effectively.

Split at random

Random Splits are used to sample a proportion of data at random and divide it into training, testing, and, ideally, validation sets. The advantage of this strategy is that the original population is likely to be well represented in all three groupings. Random splitting, to put it another way, prevents biased data sampling.

It’s crucial to remember that the validation set is used in model selection. The validation set is the second test set, and it’s understandable to wonder why there are two test sets.

The test set is used to evaluate the model during the feature selection and tuning phase. This signifies that the model parameters and feature set have been chosen to produce the best results on the test set. As a result, the validation set is utilized for the final assessment, which contains wholly unseen data points.

Cross-Validation of K-Folds

The cross-validation approach shuffles the dataset at random and then divides it into k groups. Following that, while iterating over each group, the group should be regarded as a test set, while the rest of the groups should be combined into a training set. The model is then tested on the test group, and the procedure is repeated for the remaining k groups.

As a result, at the conclusion of the procedure, one will have k distinct test group findings. The best model may then be readily chosen by selecting the model with the highest score.

K-Fold Stratified

The technique for stratified K-Fold is similar to that of K-Fold cross-validation with one major difference: unlike k-fold cross-validation, stratified k-fold considers the values of the target variable.


One of the most powerful methods for obtaining a stable model is to use Bootstrap. Because it is based on the notion of random sampling, it is similar to the random splitting approach.

The first step is to figure out how big your sample will be (which is usually equal to the size of the original dataset). After that, a random data point from the original dataset must be chosen and added to the bootstrap sample. Following the addition, the sample must be returned to the original sample. This procedure must be done N times, with N being the sample size.

As a result, the bootstrap sample is created by sampling data points from the original dataset with replacement, which is a resampling approach. This indicates that numerous occurrences of the same data point might be found in the bootstrap sample.

The model is trained on the bootstrapped sample and then tested on any data points that were not included in the bootstrapped sample. Out-of-bag samples are what they’re called.


Both model selection and model assessment procedures may appear to be complicated at first, but with practice and effective time commitment, they become second nature. Different challenges demand different approaches, and you should choose the methods that are most appropriate for your project.


Deepchecks for Computer Vision: Community Feedback Before ReleaseFeb 7th, 2022    10:00AM PST

Register Now