This package contains classification (also called as inductive inference) related programs.
In the classification problem we are given a set of training points (also called as examples). Each example consist of attributes and exactly one classification value (other names: target or function value). We assume that there exists an f probability function that returns the classification value based on the attributes. Our task is to find an f' function that well approximates (i.e. have small error) f. f' is referred to as approximating function.
Theoretically to measure the error of f' an U goodness function has to be given and we have to maximize the expected value of U(f',f). In practice we try to find a function that has the smallest misclassification rate (number of training points that are not classified properly by f').
There exists many classification procedures. The most important are: decision trees, k-nearest neighbor method, neural networks, support vector machines, bayesian networks, etc.