Generally, while performing regression, classification, or prediction, we train with 80% of the total data and test with the remaining 20%. The results are then compared with the initial training and test data. There will be some errors. These are referred to as test errors and training errors. If you are planning to get into software testing, I recommend you to check out the test architect master course from Intellipaat.
Also, if you wish to get started right away, you can watch the following video on Software Testing Tutorial to do so -