DMA

Introduction to data mining, supervised learning, unsupervised learning, and applications.

Data Mining is the extraction of interesting, non-trivial, implicit, previously unknown, and potentially useful information or patterns from data in large databases.

Statistics, Machine Learning, Database Systems, Information Visualization

Descriptive Learning and Predictive Learning

Data mining addresses the 'data explosion' problem by extracting valuable knowledge from massive amounts of data.

Database analysis, market analysis, risk analysis, fraud detection, text mining, web analysis, intelligent query answering.

Data Cleaning, Data Integration, Data Selection, Data Transformation, Data Mining, Pattern Evaluation, Knowledge Presentation.

To remove inconsistencies, eliminate noise, and handle missing values to ensure data quality.

Clustering is the task of grouping objects into clusters to maximize intra-cluster similarity and minimize inter-cluster similarity.

Customer profiling, document/image collections organization, web access pattern analysis.

Classification is a supervised learning method that assigns new observations to predefined categories based on a training dataset.

Regression is a task focused on predicting numerical output values for new objects based on known numerical values.

Scatter plots, Chernoff faces, pixel-oriented techniques, and parallel coordinates.

Data reduction simplifies datasets to make patterns more perceivable, reduce computational complexity, and speed up processing.

Data aggregation, data generalization, sampling, dimensionality reduction, quantization.

OLAP provides summarization, data slicing, and pivoting for efficient data analysis and generalization.

An automated generalization technique in data mining that simplifies data by merging and generalizing attributes.

A type of learning in data mining where a model is trained on labeled data to make predictions about new data.

Bayesian Classifiers, Linear Discriminant Functions, Support Vector Machines, Decision Trees, k-Nearest Neighbors.

By classification accuracy or error rate, model compactness, interpretability, efficiency, scalability, and robustness.

It identifies items or patterns that frequently co-occur in transaction databases to find correlations or causalities.

Descriptive learning focuses on understanding patterns, predictive learning focuses on forecasting (e.g., regression), and prescriptive learning aims to suggest actions (e.g., AI applications like autonomous driving).

Task-relevant data is data selected for its relevance to a specific data mining task, often involving feature selection, dimensionality reduction, and invariant representation.

Data integration combines data from different sources, mapping attributes (e.g., renaming C Nr to O Id) and joining tables to create a consistent dataset.

Frequent itemset mining identifies items that frequently co-occur in transactions, indicating potential correlations or causalities. Examples include market-basket analysis and association rule mining.

Basic data types include numerical (e.g., integers, floats) and categorical (e.g., symbols or identifiers).

A metric space is a set of objects with a distance function that fulfills symmetry, identity of indiscernibles, and the triangle inequality. It is useful for measuring similarity.

Composed data types include vectors, sequences, sets, and relations, often used for complex data representations.

A range query has two stages: 1) filtering to produce a candidate set using approximate distance, and 2) refining by calculating exact distances for candidates.

K-nearest neighbor queries find the closest k objects to a query object based on a specified distance metric, useful in clustering and classification tasks.

ICES criteria stand for Indexable, Complete (no false dismissals), Efficient (fast calculation), and Selective (small candidate sets) for high-quality filtering.

In multi-step search, a fast filter step first narrows down the data to candidates, followed by exact calculations on this reduced set.

Chernoff faces represent high-dimensional data as facial expressions, leveraging human intuition for similarity recognition in complex datasets.

Dimensionality reduction decreases the number of variables to improve computational efficiency and clarity, often using methods like PCA or wavelet transforms.

Equi-width histograms divide data into equal-sized ranges, while equi-height histograms contain approximately equal numbers of samples in each range, helping in data reduction.

OLAP operations like roll-up summarize data by climbing up a hierarchy, while drill-down provides more detail by moving down the hierarchy.

AOI is an automated method for generalizing data by removing or merging attributes based on specified thresholds, useful for data aggregation and simplification.

Attributes with large sets of distinct values and no generalization hierarchy, or those whose high-level concepts are covered by other attributes, may be removed.

Distributive measures combine values from partitions (e.g., count, sum), algebraic measures use a bounded number of distributive functions (e.g., average), and holistic measures lack bounded storage for representation (e.g., median).

Hierarchical indexing (e.g., R-trees) speeds up data access by organizing data spatially and pruning irrelevant data for faster query results.

The five-number summary includes the minimum, first quartile (Q1), median, third quartile (Q3), and maximum, useful for visualizing data dispersion and detecting outliers.

Scatterplot matrices help visualize pairwise correlations between multiple variables, though they are limited to showing only two dimensions at a time.

A concept hierarchy organizes values into broader categories, aiding in data generalization by structuring detailed data into meaningful abstractions.

Supervised learning uses labeled training data to create models for predicting outcomes on new data, including models like decision trees, SVMs, and k-NN.

A Bayesian classifier uses probability-based methods to classify data, useful for tasks where probabilistic predictions and uncertainty are essential.

k-NN classifies new data points based on the majority class of the k closest data points in the training set.

Classifier quality is measured by accuracy, model compactness, interpretability, efficiency, scalability, and robustness against noise or missing values.