What does error mean in machine learning?
Error (training error, validation error and test error they are all related to the same dataset): in machine learning the word error means the distance between what the model says and what is real. The model estimates an aspect of the real world, and this cannot be done with perfect accuracy. The error can be made because of noise (that is natural or unexplained variation), or because there are too bias or to variances. Having the right Bias-Variance trade-off helps to reduce the error. Bias-Variance trade-off means to find the right measure of Bias and Variance in the model. Bias is an error from assumptions made by a model and having a high-bias means that the model is really generic and not accurate/flexible, the model is underfitting. Variance is an error from sensitivity and variability inside the training data that the model tries to predict; having a high-variance, means that the model is really accurate and not generic, the model is overfitting.
K-Fold cross-validation: it is a technique to have a pretty fair level of accuracy as the test is made K time and the average performance is given as a result. Data are split into K subsets; the training data are K-1 of them, and the validation data are the last 1. This operation is repeated each time with a different combination of K-1 and 1 data.
Outliers: this is a small number of continuous numeric values that do not agree with the other because they are much smaller or bigger.
Minority values: this is the correspondent of outliers for the numeric discrete, values that rarely appear in the data.
Hyperparameter: properties of the model that controls the learning process. A hyperparameter influences the process of estimating the parameters.
Data mining: the transformation of large amounts of data into meaningful patterns and rules.