Classification is a data mining approach used to forecast team membership for data instances. It is a two-step procedure. In the first step, a model is built defining a predetermined set of data classes or approaches. The model is developed by considering database tuples defined by attributes.
Each tuple is considered to belong to a predefined class, as decided by one of the attributes, known as the class label attribute. In the framework of classification, data tuples are also defined as samples, examples, or objects. The data tuples analyzed to develop the model jointly form the training data set. The single tuples creating up the training set are defined as training samples and are casually chosen from the sample population.
Because the class label of each training sample is supported, this procedure is also referred to as supervised learning. In unsupervised learning, in which the class labels of the training samples are anonymous, and the multiple classes to be learned may not be known in advance.
The learned model is described in the structure of classification rules, decision trees, or numerical formulae. For instance, given a database of user credit data, classification rules can be learned to identify users as having either best or fair credit ratings. The rules can be used to classify future data samples, and support a good understanding of the database contents.
The holdout approach is a simple technique that applies a test set of class-labeled samples. These samples are randomly chosen and are autonomous of the training samples. The efficiency of a model on a given test set is the percentage of test set samples that are properly restricted by the model. For each test sample, the famous class label is distinguished with the learned model’s class forecast for that sample.
If the efficiency of the model were estimating depends on the training data set, this estimate can be optimistic because the learned model influence to overfit the information (namely, it can have incorporated some specific anomalies of the training information which are not present in the complete sample population). Hence, a test set is used.
Learning − Training information is analyzed by a classification algorithm. Hence, the class label attribute is a credit rating, and the learned model or classifier is described in the structure of a classification rule.
Classification − Test data are used to measure the efficiency of the classification rules. If the efficiency is treated acceptable, the rules can be used to the classification of new data tuples.