Data Classification

Algorithms and Applications

Charu C. Aggarwal

• Publisher: Chapman and Hall/CRC
• Year: 2014
• Print ISBN: 9780367659141
• eBook ISBN: 9781498760584
• Topic: Economics
• Subtopic: Statistics for Business & Economics
• Index: Economics

Comprehensive Coverage of the Entire Area of ClassificationResearch on the problem of classification tends to be fragmented across such areas as pattern recognition, database, data mining, and machine learning. Addressing the work of these different communities in a unified way, Data Classification: Algorithms and Applications explores the underlyi

Chapter 1

An Introduction to Data Classification

Charu C. Aggarwal

IBM T. J. Watson Research Center
Yorktown Heights, NY
[email protected]

1.1 Introduction

The problem of data classification has numerous applications in a wide variety of mining applications. This is because the problem attempts to learn the relationship between a set of feature variables and a target variable of interest. Since many practical problems can be expressed as associations between feature and target variables, this provides a broad range of applicability of this model. The problem of classification may be stated as follows:

Given a set of training data points along with associated training labels, determine the class label for an unlabeled test instance.

Numerous variations of this problem can be defined over different settings. Excellent overviews on data classification may be found in [39, 50, 63, 85]. Classification algorithms typically contain two phases:

• Training Phase: In this phase, a model is constructed from the training instances.
• Testing Phase: In this phase, the model is used to assign a label to an unlabeled test instance.

In some cases, such as lazy learning, the training phase is omitted entirely, and the classification is performed directly from the relationship of the training instances to the test instance. Instance-based methods such as the nearest neighbor classifiers are examples of such a scenario. Even in such cases, a pre-processing phase such as a nearest neighbor index construction may be performed in order to ensure efficiency during the testing phase.

The output of a classification algorithm may be presented for a test instance in one of two ways:

1. Discrete Label: In this case, a label is returned for the test instance.
2. Numerical Score: In this case, a numerical score is returned for each class label and test instance combination. Note that the numerical score can be converted to a discrete label for a test instance, by picking the class with the highest score for that test instance. The advantage of a numerical score is that it now becomes possible to compare the relative propensity of different test instances to belong to a particular class of importance, and rank them if needed. Such methods are used often in rare class detection problems, where the original class distribution is highly imbalanced, and the discovery of some classes is more valuable than others.

The classification problem thus segments the unseen test instances into groups, as defined by the class label. While the segmentation of examples into groups is also done by clustering, there is a key difference between the two problems. In the case of clustering, the segmentation is done using similarities between the feature variables, with no prior understanding of the structure of the groups. In the case of classification, the segmentation is done on the basis of a training data set, which encodes knowledge about the structure of the groups in the form of a target variable. Thus, while the segmentations of the data are usually related to notions of similarity, as in clustering, significant deviations from the similarity-based segmentation may be achieved in practical settings. As a result, the classification problem is referred to as supervised learning, just as clustering is referred to as unsupervised learning. The supervision process often provides significant application-specific utility, because the class labels may represent important properties of interest.

Some common application domains in which the classification problem arises, are as follows:

• Customer Target Marketing: Since the classification problem relates feature variables to target classes, this method is extremely popular for the problem of customer target marketing. In such cases, feature variables describing the customer may be used to predict their buying interests on the basis of previous training examples. The target variable may encode the buying interest of the customer.
• Medical Disease Diagnosis: In recent years, the use of data mining methods in medical technology has gained increasing traction. The features may be extracted from the medical records, and the class labels correspond to whether or not a patient may pick up a disease in the future. In these cases, it is desirable to make disease predictions with the use of such information.
• Supervised Event Detection: In many temporal scenarios, class labels may be associated with time stamps corresponding to unusual events. For example, an intrusion activity may be represented as a class label. In such cases, time-series classification methods can be very useful.
• Multim...

Table of contents

• Preliminaries
• Series
• Dedication
• Editor Biography
• Contributors
• Preface
• Chapter 1 An Introduction to Data Classification
• Chapter 2 Feature Selection for Classification: A Review
• Chapter 3 Probabilistic Models for Classification
• Chapter 4 Decision Trees: Theory and Algorithms
• Chapter 5 Rule-Based Classification
• Chapter 6 Instance-Based Learning: A Survey
• Chapter 7 Support Vector Machines
• Chapter 8 Neural Networks: A Review
• Chapter 9 A Survey of Stream Classification Algorithms
• Chapter 10 Big Data Classification
• Chapter 11 Text Classification
• Chapter 12 Multimedia Classification
• Chapter 13 Time Series Data Classification
• Chapter 14 Discrete Sequence Classification
• Chapter 15 Collective Classification of Network Data
• Chapter 16 Uncertain Data Classification
• Chapter 17 Rare Class Learning
• Chapter 18 Distance Metric Learning for Data Classification
• Chapter 19 Ensemble Learning
• Chapter 20 Semi-Supervised Learning
• Chapter 21 Transfer Learning
• Chapter 22 Active Learning: A Survey
• Chapter 23 Visual Classification
• Chapter 24 Evaluation of Classification Methods
• Chapter 25 Educational and Software Resources for Data Classification