Cross Validation

Use case

Cross Validation is a model validation technique used to check how well a model generalizes by checking how it performs against new unseen data.

Types

1. Leave One Out Cross Validation
   • A datum is removed from the the data and the model is trained on it.
   • Repeat this for each datum and average the errors.
   • Thus a truly unbiased error of the model is obtained.
   • However, not feasible.
2. K-fold Cross Validation
   • It is an alternative to Leave One Out method and provides a good approximation.
   • Here instead of individually removing a datum and training the remaining the data, the data set is split into $(k-1)$ Training set with one Test set.
   • The error is computed for each set and averaged.
   • The whole process is the repeated with a different Test Set.