Untitled Quiz

Data Mining

• Data mining extracts or mines knowledge from large data sets.
• It's a computational process finding patterns in large datasets using methods from artificial intelligence, machine learning, statistics, and database systems.
• The aim is to extract information from data and turn it into a usable structure.
• Key properties include automatic pattern discovery, prediction of outcomes, creation of actionable information, and a focus on large datasets.
• It draws similarities from searching for valuable business information in large databases, like store scanner data, and finding valuable ore in a mountain.

Scope of Data Mining

• Data mining's name reflects its similarity with searching for valuable business information in large databases or mining a mountain for valuable ore.
• Databases of sufficient size and quality allow for data mining.

Tasks of Data Mining

• Anomaly detection (outlier/change/deviation detection) identifies unusual data.
• Association rule learning finds relationships between variables (e.g., supermarket basket analysis).
• Clustering discovers groups of similar data points.
• Classification generalizes known structure to new data (e.g., spam detection).
• Regression attempts to model data with the least error.

Architecture of Data Mining

• A typical data mining system has several components.
• Knowledge base: Domain knowledge guides the search and evalutes interestingness of patterns.
• Data mining engine: Processes mining tasks like characterization, association analysis, classification, prediction, cluster analysis, outlier analysis, and evolution analysis.
• Pattern evaluation module: Uses interestingness measures to focus the search on interesting patterns. Filtering is also possible.
• User interface: Communicates between users and the system for tasks like query input and result visualization.

Data Mining Process

• State the problem and formulate hypothesis. Domain knowledge is crucial for a meaningful problem statement. Several hypotheses might be formulated for a problem. Data mining expert and application expert collaboration is needed.
• Collect the data. Data generation can be performed by an expert (designed experiment) or not influenced by an expert (observational approach).
• Preprocess the data. This step involves cleaning, transforming, and integrating data.
  • Data cleaning: Identifies and corrects data errors (missing values, outliers, duplicates). Techniques include imputation, removal, and transformation.
  • Data integration: Combines data from multiple sources into a unified data set. Techniques include record linkage and data fusion.
  • Data transformation: Converts data into a suitable format for analysis. Techniques include normalization, standardization, and discretization.
  • Data reduction: Reduces data size while preserving valuable information. Techniques include feature selection, feature extraction, sampling, and clustering.
• Estimate the model. Selection and implementation of the appropriate data mining technique is needed. Several models and selecting the best is an additional task.
• Interpret the model and draw conclusions. The data mining model assists in decision making; hence, its interpretation is important.

Knowledge Discovery in Databases (KDD)

• Knowledge discovery is a process of discovering patterns and derived values from data. Several steps are involved as follows:
  • Data cleaning.
  • Data integration.
  • Data selection.
  • Data transformation.
  • Data mining.
  • Pattern evaluation.
  • Knowledge presentation.

Data Warehouse

• A data warehouse is a subject-oriented, integrated, time-variant, and non-volatile collection of data.
• Subject-oriented: Used for analyzing a particular subject area (e.g., sales).
• Integrated: Integrates data from multiple sources into a singular representation.
• Time-variant: Stores historical data (e.g., over months or years).
• Non-volatile: Data cannot be altered after it's stored in the data warehouse.

Data Warehouse Design Process

• Top-down approach: Starts with overall design and planning, useful if technology is mature.
• Bottom-up approach: Starts with experiments and prototypes, useful in the early stage of development.
• Combined approach: Combines both top-down and bottom-up strategies.

Three Tier Data Warehouse Architecture

• Bottom tier: Data warehouse database server (relational database system).
• Middle tier: OLAP server (relational OLAP (ROLAP) or multidimensional OLAP (MOLAP)).
• Top tier: Front end client layer with query and reporting tools, analysis tools, and/or data mining tools.

Data Warehouse Models

• Enterprise warehouse: Collects all information from an organization. Scope is corporate-wide.
• Data mart: A subset of data which is of value for a specific group of users. The scope is confined.
• Virtual warehouse: A set of views over operational databases.

Meta Data Repository

• Metadata are data about data (in a data warehouse).
• Includes data lineage, currency, and monitoring information.

OLAP (Online Analytical Processing)

• OLAP is an approach to answering multi-dimensional analytical queries swiftly.
• Key operations are consolidation, drill-down, slicing, and dicing
• Helps with analysis of data based on several perspectives, such as the time, location, and product type of a sale.

Data Cube

• A data cube is a multi-dimensional structure for representing data summarization.
• It facilitates fast and efficient analysis of data across dimensions.
• Dimensions determine how data is summarized. Facts are measures used to analyze data relationships.

Data Cube Computation

• Effective optimization techniques are used in computation for efficient tasks like sorting, hashing, and grouping.
• Caching intermediate results is a common method to improve performance.

Data Generalization

• Transforming raw data into a more simplified form for analysis.
• This simplifies data analysis and identification of patterns. Example techniques are clustering, sampling, and dimensionality reduction.

Data Cube Approach

• A method used to handle large quantities of data efficiently by creating a multi-dimensional structure called the data cube.
• Summarization of data along various dimensions is done for quick query processing and analysis.

Data Attribute Orientation Induction

• Another method for data generalization by creating rules as attribute orientations.
• Used to classify data points based on their characteristics.

Frequent Pattern Mining

• Identifying frequently occurring patterns in data.
• Methods are used depending on the kind of patterns sought.
• Examples of these techniques include association rule mining and sequential pattern mining.

Efficient Frequent Itemset Mining Methods

• Apriori algorithms are a frequently used technique for frequent itemset mining due to their efficiency, using prior knowledge of frequent itemset properties

Bayesian Classification

• A statistical classification method based on Bayes' theorem.
• Predicts probabilities for class-membership which can be used for tasks where accuracy is not critical.

Multilayer Feed-forward Neural Network

• A neural network type where data moves through layers.
• Backpropagation algorithm is a training method using gradient descent for iterative processing of training tuples.

k-Nearest-Neighbor Classifier

• Used to classify a new data point based on analogy.
• Finds the k most similar existing data points and classifies similarly.

Support Vector Machine

• A technique to create a hyperplane for classification using extreme points.
• These points that lie closest to the optimal hyperplane are called support vectors. They are extremely important to the method's accuracy because they determine the hyperplane's location.

Cluster Analysis

• Clusters are groups of similar data objects.
• Clustering algorithms are used to find these similarities.
• Cluster analysis includes different approaches.

Partitioning Methods

• Dividing data objects into groups.

Hierarchical Methods

• Construct a hierarchical decomposition of the data.

Density-Based Methods

• Clustering methods based on the density of the data points in a neighborhood.

Grid-Based Methods

• Quantizing object space into a grid structure for fast processing.

Model-Based Methods

• Hypothesizing a model for each cluster in the dataset.
• Clusters are given by a density distribution.

Clustering High-Dimensional Data

• Handling many features in clustering.

Constraint-Based Clustering

• Clustering which adapts to user-specified conditions or constraints

Classical Partitioning Methods

• Methods such as k-means clustering are common for dividing data into groups of k clusters where items within the cluster are close and items between clusters are far.

k-Medoids Method

• k-means alternative based on medoids (or central objects). It is more robust to outliers than the k-means method.

Hierarchical Clustering Methods

• hierarchical methods create a tree, or hierarchy, of clusters. There are agglomerative and divisive approaches to building these hierarchies.

Constraints Based Clustering Analysis

• The specifications of constraints will affect the methodology used for the process. Several types of constraints are common; like ones for object selection, cluster parameter selection, or the distance functions for objects.

Outlier Analysis

• Finding data points that deviate from the general patterns or behaviors of the data set.

Statistical Distribution Based Outlier Detection

• Identify objects that do not comply with the general data pattern utilizing the hypothesis based approach.

Distance Based Outlier Detection

• Identify data points that are far from other points in the data set using a distance metric.

Density Based Local Outlier Detection

• A method to identify outliers that may not fit into a homogenous cluster shape.

Deviation Based Outlier Detection

• A method identifying objects that deviate from the main characteristics of the group they belong to.