In machine learning, LOOCV stands for leave-one-out cross validation. Assuming that the dataset of an ML project comprises n examples (rows), we split the dataset into n subsets and perform n iterations of training/testing. In each iteration, n-1 examples are used as training data and only 1 example is used the testing subset. This process is repeated n times. The LOOCV data validation method usually leads to low bias (since all the dataset's data is used for training) but also high variance (since using only a single subset each time can potentially be an outlier). Using another cross validation method, such as k-fold cross validation, may be more appropriate in certain scenarios. Essentially, loocv is a case of k-fold cross validation where k=n, where n is the size of the dataset.
