Transcript Data Mining
An Introduction to Data Mining Hosein Rostani Alireza Zohdi Report 1 for “advance data base” course Supervisor: Dr. Masoud Rahgozar December 2007 1 Outline Why data mining? Data mining applications Data mining functionalities Concept description Association analysis Outlier Analysis Evolution Analysis Classification Clustering 2 Why data mining? Motivation: Wide availability of huge amounts of data Need for turning data into useful info & knowledge Data mining: Extracting or “mining” knowledge from large amounts of data Knowledge : useful patterns Semiautomatic process Focus on automatic aspects 3 Data mining applications Prediction. Examples: Credit risk Customer switching to competitors Fraudulent phone calling card usage Associations. Examples: Related books for buy Related accessories for suggest: e.g. camera Causation discovery: e.g. medicine Clusters. Example: Clusters of disease 4 Data mining functionalities Concept description Characterization & discrimination Association analysis Outlier Analysis Evolution Analysis Classification and Prediction Clustering 5 Concept description Description of concepts summarized, concise & precise Ways: Data characterization Summarizing the data of the target class in general terms Data discrimination Comparison of the target class with the contrasting class(es) Examples of Output forms: Pie charts, bar charts, curves & multidimensional tables 6 Association analysis Mining frequent patterns For discovery of interesting associations within data Kinds of frequent patterns: Frequent itemset Set of items frequently appear together. E.g. milk and bread Frequent subsequence E.g. pattern of customers’ purchase: First a PC, then a digital camera & then a memory card Frequent substructure Structural forms such as graphs, trees, or lattices Support and confidence 7 Outlier Analysis Outliers: data objects disobeying the general behavior of data Approaches to outliers Discard as noise or exceptions Keep for applications such as fraud detection Example: detecting fraudulent usage of credit cards Ways: Using statistical tests Using distance measures Using deviation-based methods 8 Evolution Analysis Description and modeling of trends For objects with changing behavior over time Ways: Applying other data mining tasks on time related data Association analysis, classification, prediction, clustering & … Distinct ways time-series data analysis sequence or periodicity pattern matching similarity-based data analysis Example: stock market: predict future trends in prices 9 Classification and Prediction Classification: Process of finding a model that distinguishes data classes Purpose: using the model to predict the class of new objects Deriving model: Based on the analysis of a set of training data data objects with known class labels Example: In a credit card company Classification of customers based on their payment history 10 Classification A two-step process for classification: First: Learning or training step Building the classifier by analyzing or learning from training data Second: classifying step Using classifier for classification Accuracy of a classifier (on a given test set) Percentage of test set tuples correctly classified by classifier Classification methods: Decision tree, Naïve Bayesian classification, Neural network, k-nearest neighbor classification, … 11 Decision tree Decision tree induction : Learning of decision trees from class-labeled training tuples Decision tree: A flowchart-like tree structure Internal nodes: tests on attributes Branches: outcomes of the test Leaves: class labels Usage in classification: Prediction by tracing a path from the root to a leaf node Testing attribute values of new tuple against decision tree Easily converting Decision tree to classification rules 12 Decision tree example: Does a customer buys a computer? 13 Bayesian Classification Bayesian classification Predicting the probability that a new tuple belongs to a particular class High accuracy and speed in large databases Based on Bayes’ theorem Conditional probability Naïve Bayesian classifier Assumption: class conditional independence Good for Simplifying computations 14 Clustering The process of grouping a set of physical or abstract objects into classes of similar objects Generating class labels for objects currently without label Clustering based on this principle: Maximizing the intraclass similarity and Minimizing the interclass similarity Clustering also for facilitating taxonomy formation Hierarchical organization of observations 15 An example: clustering customers in a restaurant Restaurant database Preprocessing Object View for Clustering Clustering A Set of Similar Object Clusters Summarization Young at midnight White Collar for Dinner Retired for Lunch 16 Steps of database Clustering 1. 2. 3. 4. 5. 6. 7. Define object-view Select relevant attributes Generate suitable input format for the clustering tool Define similarity measure Select parameter settings for the chosen clustering algorithm Run clustering algorithm Characterize the computed clusters 17 Challenge: database clustering Data collections are in many different formats Flat files Relational databases Object-oriented database Flat file format: The simplest and most frequently used format in the traditional data analysis area Databases are more complex than flat files 18 Challenge: database clustering (cont.) Challenge: Changing clustering algorithms to become more directly applicable to real-world databases Issues related to databases: Different types of objects in DB Relationships between objects: 1:1, 1:n & n:m Complexity in definition of object similarity Due to the presence of bags of values for an object Difficulty in selection of an appropriate similarity measure Due to the presence of different types for attributes of objects 19 Refferences Han, J., Kamber, M., Data Mining: Concepts and Techniques, Second Edition, Elsevier Inc., 2006, 770 p., ISBN 1-55860-901-3. Silberschatz, A., Korth, F., Sudarshan, S., Database System Concepts, Fifth Edition, McGraw-Hill, 2005, ISBN 0-07-295886-3. Ryu, T., Eick, C., A Database Clustering Methodology and Tool, in Information Sciences 171(1-3): 29-59 (2005). 20