Classification and prediction with stages

Classification is a supervised learning task that predicts a discrete class label for an object using a model learnt from training data with known labels. Example: classifying a loan applicant as "safe" or "risky".

Prediction estimates a continuous or missing numeric value (regression). Example: predicting how much a customer will spend on a computer. (Some authors use "prediction" for both.)

Stages of classification

1. Data preparation: clean the data, select relevant attributes, normalise and discretize, and divide the data into a training set and a test set.
2. Model construction (learning step): a classification algorithm (decision tree, Naive Bayes, neural network) analyses the training tuples, whose class labels are given, and builds a model such as rules, a tree or a formula.
3. Model evaluation: the model is applied to the independent test set, and accuracy = correctly classified/total is found. If it is not acceptable, the model is rebuilt with other parameters or algorithms.
4. Model usage (classification step): the accepted model classifies new unseen tuples.

Training data --> [Learning algorithm] --> Model
                                            |
Test data ------> [Evaluate accuracy] <-----+
                          | acceptable?
New data  ------> [Use the model] ---> predicted class

Example: Training tuples (name, rank, years, tenured) are used to build the rule IF rank = professor OR years > 6 THEN tenured = yes. After the test accuracy is found, the rule classifies the new tuple (Jeff, professor, 4).