Transcript Data Mining Techniques ACM-SIGMOD`96 Conference Tutorial

Spatial Data Mining and
Spatial Data Warehousing
Special Topics In Database
Sadra Abedinzadeh
Ashkan Zarnani
Farzad Peyravi
Outline

Motivation and General Description

Data Warehousing: Basic Concepts and Techniques

Spatial Data Warehousing and Spatial OLAP Techniques
– Spatial Data Warehouse: Models and Construction
– Spatial OLAP: Implementation and Application

Data Mining: Basic Concepts and Techniques

Spatial Data Mining
– Mining Spatial Association Rules.
– Spatial Classification and Prediction
– Spatial Data Clustering Analysis

Conclusions and Future Research.
Motivation

Data warehousing: Integrating data from multiple sources
into large warehouses and support on-line analytical
processing and business decision making.

Data mining (knowledge discovery in databases):
Extraction of interesting knowledge (rules, regularities,
patterns, constraints) from data in large databases.

Necessity: Data explosion problem --- computerized data
collection tools and mature database technology lead to
tremendous amounts of data stored in databases.

We are drowning in data, but starving for knowledge!
Data Warehousing

“ A data warehouse is a subject-oriented, integrated, timevariant, and nonvolatile collection of data in support of
management’s decision-making process.” --- W. H. Inmon

A data warehouse is
– A decision support database that is maintained
separately from the organization’s operational databases.
– It integrates data from multiple heterogeneous sources to
support the continuing need for structured and /or ad-hoc
queries, analytical reporting, and decision support.
Modeling Data Warehouses


Modeling data warehouses: dimensions & measurements
– Star schema: A single object (fact table) in the middle
connected to a number of objects (dimension tables)
radially.
– Snowflake schema: A refinement of star schema where
the dimensional hierarchy is represented explicitly by
normalizing the dimension tables.
– Fact constellations: Multiple fact tables share dimension
tables.
Storage of selected summary tables:
– Independent summary table storing pre-aggregated data,
e.g., total sales by product by year.
– Encoding aggregated tuples in the same fact table and the
same dimension tables.
Example of Star Schema
Time Dimension Table
Sales Fact Table
Product Dimension Table
Many Time Attributes
Time_Key
Many Product Attributes
Product_Key
Store Dimension Table
Many Store Attributes
Store_Key
Location_Key
unit_sales
Measurements
dollar_sales
Yen_sales
Location Dimension Table
Many Location Attributes
Example of a Snowflake Schema
Supplier_Key
Time Dimension Table
Many Time Attributes
Store Dimension Table
Many Store Attributes
Sales Fact Table
Product Dimension Table
Time_Key
Supplier_Key
Product_Key
Product_Key
Store_Key
Location_Key
Location Dimension Table
Location_Key
unit_sales
Measurements
dollar_sales
Country
Location_Key
Yen_sales
Region
Location_Key
A Star-Net Query Model
Customer Orders
Shipping Method
Customer
CONTRACTS
AIR-EXPRESS
ORDER
TRUCK
PRODUCT LINE
Time
Product
ANNUALY QTRLY
DAILY
PRODUCT ITEM PRODUCT GROUP
DISTRICT
SALES PERSON
REGION
DISTRICT
COUNTRY
DIVISION
Geography
Promotion
Organization
Construction of Data Cubes
Amount
B.C.
Province Prairies
Ontario
sum
0-20K20-40K 40-60K60K- sum
All Amount
Comp_Method, B.C.
Comp_Metho
d
Database
… ...
Discipline
sum





Each dimension contains a hierarchy of values for one attribute
A cube cell stores aggregate values, e.g., count, sum, max, etc.
A “sum” cell stores dimension summation values.
Sparse-cube technology and MOLAP/ROLAP integration.
“Chunk”-based multi-way aggregation and single-pass computation.
Efficient Data Cube Computation Methods

Data cube can be viewed as a lattice of cuboids
– The bottom-most cuboid is the base cube.
– The top most cuboid contains only one cell.

Materialization of data cube
– Materialize every (cuboid), none, or some.
– Algorithms for selection of which cuboidsALL
to materialize.
 Based

on size, sharing, and access frequency.
B
Efficient cube computation methods
A
C
– ROLAP algorithms.
AB
– Array-based cubing algorithm.
BC
ABC
AC
OLAP: On-Line Analytical
Processing

A multidimensional, LOGICAL view of the data.

Interactive analysis of the data: drill, pivot, slice_dice, filter.

Summarization and aggregations at every dimension
intersection.

Retrieval and display of data in 2-D or 3-D crosstabs, charts,
and graphs, with easy pivoting of the axes.

Analytical modeling: deriving ratios, variance, etc. and
involving measurements or numerical data across many
dimensions.

Forecasting, trend analysis, and statistical analysis.

Requirement: Quick response to OLAP queries.
OLAP Architecture



Logical architecture:
– OLAP view: multidimensional and logic presentation of
the data in the data warehouse/mart to the business user.
– Data store technology: The technology options of how and
where the data is stored.
Three services components:
– data store services
– OLAP services, and
– user presentation services.
Two data store architectures:
– Multidimensional data store: (MOLAP).
– Relational data store: Relational OLAP (ROLAP).
Spatial Data Warehouse and Spatial OLAP



Spatial Data Warehouse: Integrated, subject-oriented,
time-variant, and nonvolatile spatial data repository for
data analysis and decision making.
Spatial Data Integration: A big issue.
Spatial data cube: Multidimensional spatial database.
– Non-spatial dimensions: time, product, organization
hierarchies.
– Spatial dimensions: formed by geo-spatial hierarchies.
– Non-spatial (numerical) measurements:
 Distributive, algebraic, holistic.
– Spatial Measurements:
 Collection of spatial object pointers which may
require spatial merge, overlay, or other operations.
Example: Weather Pattern Analysis




Input:
– a map with about 3,000 weather probes scattered in B.C.
– daily data for temperature, precipitation, wind velocity,
etc.
– concept hierarchies for all attributes
Output:
– a map that reveals patterns: merged (similar) regions!
Goals:
– interactive analysis (drill-down, slice, dice, pivot, roll-up)
– fast response time
– minimizing storage space used
Challenge: a merged region may contain hundreds of
“primitive” regions (polygons).
A Model of Spatial Data Warehouses

Dimensions
 Measurements
– nonspatial
– numerical
 (e.g. 25-30 degrees
 distributive (e.g. count,
generalizes to hot)
sum)
– spatial-to-nonspatial
 algebraic (e.g. average)
 (e.g. region “B.C.”
 holistic (e.g. median,
generalizes to
rank)
description “western
– spatial
provinces”)
 collection of spatial
– spatial-to-spatial
pointers (e.g. pointers to
 (e.g. region “Burnaby”
all regions with 25-30
generalizes to region
degrees in July)
“Lower Mainland”)
Star Model of a Spatial Data
Warehouse

Dimensions
– region_name
– time
– temperature
– precipitation

Measurements
– region_map
– area
– count
Dimension table
Fact table
Spatial Merge: Pre- vs On-line
Computation
Precomputing all: too much storage space
On-line merge:
very expensive
Spatial Measurements: Selective
Materialization


Methods for computation of spatial measurements in spatial
data cube.
– Collect and store pointers to spatial objects in a spatial
data cube:Computing on the fly --- expensive and slow.
– Saving all the possible combinations --- huge space
overhead.
– Precompute and store rough approximations in a spatial
data cube --- accuracy trade-off.
– Selective computation: only materialize those which will
be accessed frequently --- a reasonable choice.
Cube lattice and granularity of merge-able spatial objects.
– Cuboid-level vs. cube cell level granularity.
Computing Spatial Measurements


Apply [HRU96] greedy
algorithm to select cuboids
[HRU96] algorithm has
granularity on a cuboid level
Region_name
Northern BC
Interior BC
Kooteney
Lower Main.
Vanc Isl.
Okanagan




Finer granularity, on a cell level
cold mod warm hot Temperature
Only selected cells are materialized (not the whole cuboid)
Factors in selections of cells
– access frequency
– size of a cell (number of merged objects)
It could be better to save {1,3,4,7} than {1,3}
– benefit for on-the-fly computation:If {1,3} is saved, it can be
used for {1,3,6}.
Only neighboring objects are merged.
Integration of Data Mining and
Data Warehousing

Data warehouse provides clean, integrated data for fruitful
mining.

Data mining provides powerful tools for analysis of data
stored in data warehouses.

OLAP can be viewed as data summarization and simple
data mining.

Data mining provides more analysis tools, e.g., association,
classification, clustering, pattern-directed, and trend
analysis.

Mining multi-level knowledge by integration with OLAP
facilities: mining in multiple data cubes.
Mining Different Kinds of
Knowledge

Characterization: Generalize, summarize, and possibly
contrast data characteristics, e.g., dry vs. wet regions.

Association: Rules like “inside(x, city)  near(x, highway)”.

Classification: Classify data based on the values in a
classifying attribute, e.g., classify countries based on climate.

Clustering: Cluster data to form new classes, e.g., cluster
houses to find distribution patterns.

Trend and deviation analysis: Find and characterize evolution
trend, sequential patterns, similar sequences, and deviation
data, e.g., housing market analysis.

Pattern-directed analysis: Find and characterize userspecified patterns in large databases, e.g., volcanos on Mars.
Different Mining Tasks in Spatial DBs


Spatial data mining tasks:
– Spatial data characterization and comparison
– Spatial clustering analysis
– Spatial classification
– Spatial association
– Spatial pattern analysis
Spatial concept hierarchies: thematic vs. spatial.
– Thematic hierarchy: e.g., agriculture (food (grain (corn, rice,
...), vegetable, fruit), others(...)).
– Spatial hierarchy, based on
 Spatial data structures (MBR, quad-tree & R-tree).
 Spatial related semantics (geo-region classification).
 Clustering analysis (e.g., neighborhood or adjacent_to).
A Geo-Spatial Data Mining Query
Language: GMQL
Extension to Spatial SQL [Egenhofer’94].
Support ad-hoc data mining queries.
mine characteristic rules
type of rule (characteristic, discriminant,
association, clustering, classification)
for “Description of states along I 80 highway”
from us_hiway, states_census
SQL like from, where clauses
where states_census.obj intersects us_hiway.obj high level concepts and
and highway = "I 80”
spatial joins may be used
with respect to states_census.obj, state_name, pop90, capita_income
list of relevant attributes
set attribute threshold 51 for state_name thresholds for rules filtration
Background Knowledge for Data
Mining




Conceptual "hierarchies" and generalization operators.
– Instance-based: {freshman, ..., senior} undergraduate.
– Schema-based: address(city, province, country).
– Rule-based: good(x)  undergraduate(x)  gpa(x)  3.5.
– Operation-based: aggregation, approximation, clustering, etc.
Where to get such background knowledge?
– Implicitly stored in databases, such as address.
– Explicitly defined by experts, such as "physics science".
– Formed with different attribute combinations,
 food(category, brand, content _spec, package _size, price).
– Generated automatically by data distribution analysis.
May need dynamic adjustment for a particular set of data.
Choose from multiple hierarchies or try them in parallel.
Automatic Generation of Numeric Hierarchies
Count40
35
30
25
20
15
10
5
Amount
0
10000
30000
50000
70000
90000
2000-97000
2000-16000
2000-12000
16000-97000
12000-16000
16000-23000
23000-97000
Spatial OLAP (Characterization)
Viewing data from different angles
 Summarization on multiple concept levels

Spatial
slicing
Drillingdown on
medium
family
income
Mining Discriminant Rules



Discrimination: Comparison of two or more classes
Strategy:
– Collect the relevant data respectively into the target
class and the contrasting class
– Generalize both classes to the same high level concepts,
– Compare tuples with the same high level descriptions,
– Present for every tuple its description and two numbers
support - distribution within single class
comparison - distribution between classes
– Highlight the tuples with strong discriminant features
Interestingness:
– Different measures of interestingness,
e.g. consider also the sizes of different classes
Spatial OLAP (Comparison)

Comparing different classes of data
Population increases faster in the
western part.
Drill down, and look at different
dimensions to get explanation!!
Mining Association Rules

Association: Finding association among a set of attributes
and their values.

Applications: pattern association, market analysis, etc.

Examples.
– milk bread [5%, 60%]
– tire  auto_accessories  auto_services [2%, 80%]

Methods for mining associations :
– Apriori ( Agrawal & Srikant’94)
– Partition technique (Savasere, Omiecinski, Navathe’95)
– Sampling (Toivonen’96)
Spatial Associations
FIND SPATIAL ASSOCIATION RULE DESCRIBING
"Golf Course"
FROM Washington_Golf_courses, Washington
WHERE CLOSE_TO(Washington_Golf_courses.Obj,
Washington.Obj, "3 km")
AND Washington.CFCC <> "D81"
IN RELEVANCE TO Washington_Golf_courses.Obj,
Washington.Obj, CFCC
SET SUPPORT THRESHOLD 0.5
Spatial Associations & Hierarchy of
Spatial Relationships


Spatial association: Association relationship containing
spatial predicates, e.g., close_to, intersect, contains, etc.
– Topological relations:
 intersects, overlaps, disjoint, etc.
– Spatial orientations:
 left_of, west_of, under, etc.
– Distance information:
 close_to, within_distance, etc.
Hierarchy of spatial relationship:
– “g_close_to”: near_by, touch, intersect, contain, etc.
– First search for rough relationship and then refine it.
Efficient Mining of Spatial Associations

Two-step computation of spatial associations:
– Step 1: rough spatial computation as a filter
 MBR
or R-tree rough estimation.
– Step2: Detailed spatial algorithm as refinement
 apply
only to those pairs which have passed the rough
spatial association testing (no less than min_support).

Multi-dimensional mining:
– explore association relationships at any selected
granularity level
– perform drill-down and roll-up on any dimension.
Example: Spatial Association Rule Mining

“What kinds of spatial objects are close to each other in
B.C.?”
– Kinds of objects: cities, water, forests, usa_boundary,
mines, etc.

Rules mined:
– is_a(x, large_town) ^ intersect(x, highway) 
adjacent_to(x, water). [7%, 85%]
– is_a(x, large_town) ^adjacent_to(x, georgia_strait) 
close_to(x, u.s.a.). [1%, 78%]

Mining method: Ariori + multi-level association + geo- spatial
algorithms (from rough to high precision).
Data Classification



Data categorization based on a set of training objects.
– Applications: credit approval, target marketing,
medical diagnosis, treatment effectiveness analysis, etc.
– Example: classify a set of diseases and provide the
symptoms which describe each class or subclass.
The classification task: Based on the features present in
the class_labeled training data, develop a description or
model for each class. It is used for
– classification of future test data,
– better understanding of each class, and
– prediction of certain properties and behaviors.
Data classification methods: Decision-trees (e.g., ID3,
C4.5), statistics, neural networks, rough sets, etc.
A Decision-Tree Based Classification Method

A decision tree:
outlook
sunny
humidity
N

P
rain
overcast
windy
P
N
P
ID-3 and C4.5 (Quinlan’93): A top-down decision tree
generation algorithm.
– At start, all the training examples are at the root.
– Partition examples recursively based on selected
attributes.
– Attribute selection: Maximizing an information gain
measure, i.e., favoring the partitioning which makes
the majority of examples belong to a single class.
Scalable Classification Methods



Scalability of decision-tree classification algorithms.
Previous approaches:
– Incremental tree construction (Quinlan’86): total cost is high.
– Data sampling and discretizing continuous attributes
(Cattlet’91): still in main memory.
– Data partition and merge of parallel partition (Chan and
Stolfo’91): reduced classification accuracy.
SLIQ & SPRINT (Mehta et al.’96, Shafer et al.’96): disk-based
– Decision-tree construction algorithms.
– Techniques: Pre-sorting, breadth_first tree-growing, and treepruning.
Generalization-Based Decision-Tree Induction




Integration of generalization with decision-tree induction.
Classification at primitive concept levels, e.g., precise
temperature, humidity, outlook, etc.
– Weakness: low-level concepts, scattered classes, bushy
classification-trees, semantic interpretation problems.
Classification at high or medium concept levels:
– may lead to imprecise classification.
Medium level generalization & adjustment:
– Generalize to intermediate concept level(s).
– Merge and split concept levels for better class
representation and classification accuracy.
– Efficiency: Analysis performed in compressed, generalized
relations.
Mining Classification Rules
 Classification: Based on the
features present in the
class_labeled training data,
develop a description or
model for each class.
 Applications: credit approval,
target marketing, medical
diagnosis, treatment
effectiveness analysis, etc.
 Example: classify a set of
diseases and provide the
symptoms which describe
each class or subclass.
Spatial Classification
Generalization-based
induction
 Interactive
classification

Predictive Modeling in Databases





Predictive modeling: Predict data values or construct
generalized linear models based on the database data.
One can only predict value ranges or category distributions.
Method outline:
– Minimal generalization
– Attribute relevance analysis
– Generalized linear model construction
– Prediction.
Determine the major factors which influence the prediction.
– Data relevance analysis: uncertainty measurement, entropy
analysis, expert judgement, etc.
Multi-level prediction: drill-down and roll-up analysis.
Spatial Prediction and Trend
Analysis



Spatial trend predictive modeling (Ester et al’97):
– Discover centers: local maximal of some non-spatial
attribute.
– Determine the (theoretical) trend of some non-spatial
attribute, when moving away from the centers.
– Discover deviations (from the theoretical trend).
– Explain the deviations.
Example: Trend of unemployment rate change
according to the distance to Munich.
Similar modeling can be used to study trend of
temperature with the altitude, degree of pollution in
relevance to the regions of population density, etc.
Data Clustering Analysis
Data clustering (“unsupervised learning”): Cluster objects
into classes, based on their features, which maximize
intraclass similarity and minimize interclass similarity.
 Probability-based vs. distance-based clustering analysis.
 Typical probability-based clustering analysis algorithms:
– COBWEB (Fisher’87): Incremental concept formation.
 Category utility measurement (probability of each
concept’s occurrence)
 Top-down, incremental, hierarchical organization of
concepts.
– CLASSIT (Gennari’89): extend it to real-valued data.
 Typical distance-based clustering analysis algorithms:
– Statistics-based, k-means, k-medoids, nearest neighbors.

Distance-Based Spatial Clustering
Analysis
Statistical approaches: scan data frequently, iterative
optimization, hierarchical clustering, etc.
 CLARANS (Ng & Han’94): randomized search (sampling)
+ PAM (a distance-based clustering algorithm).
 DASCAN (Ester et al.’96): density-based clustering using
spatial data structures (R*-tree).
 BIRCH (Zhang et al.’96): Balanced iterative reducing and
clustering using hierarchies.
– Focus on densely occupied portions of the data space.
– Measurement reflects the “natural” closeness of points.
– A height-balanced tree (CF-tree) is used for clustering.
 Describe aggregate proximity relationships (Knorr & Ng’96).

Spatial Clustering
How can we cluster points?
 What are the distinct features of the clusters?

There are more customers with university degrees
in clusters located in the West.
Thus, we can use different marketing strategies!
Data and Knowledge Visualization







Visualization of characteristic and discriminant rules:
– tables & cubes + bar/pie charts, curves, surfaces, etc.
Visualization of association rules:
– Association rule graph: Nodes for large 1-itemset, lines for
large 2-items sets, arrows for implication strength.
– Association matrix: support/confidence: size/color in cells.
Cluster analysis: viewing clusters and their characteristics.
Classification: colored decision trees.
Prediction: curves, pie charts, and relevance analysis results.
Deviation analysis: boxplots (quartiles, median) and outliers.
Visual impression of large data mining results
– arrange and color data items as pixels (Keim et al.’94)
Visual Data Mining (ref. D. Keim SIGMOD’96




Tutorial)
Data visualization and exploratory analysis:
– Interactive, usually undirected search for structures,
trends, etc.
Typical data visualization techniques:
– Geometric techniques, icon-based techniques, pixeloriented techniques, hierarchical techniques, graph-based
techniques, 3D-techniques, dynamic techniques, and
hybrid techniques.
Database visualization systems:
– Statistics-oriented systems, visualization-oriented
systems, database-oriented systems and special purpose
systems.
Visual database exploration is another powerful approach to
data mining, especially spatial data mining.
Data Mining Interfaces



Interactive mining versus a data mining language.
Specification of data mining tasks.
– Data sets: any sets of data in databases
– Mining task specification: kinds of knowledge or forms of
rules to be mined.
– Background knowledge (e.g., concept hierarchies):
specification and manipulation.
– Interestingness measurement: significance, confidence,
thresholds, concept levels, etc.
Transformation and manipulation of output results.
– Roll-up vs. drill-down.
– Multiple output forms: generalized relations, crosstabs,
charts, curves, and other visual outputs.
GeoMiner: Graphical User
Interface
Systems for Data Warehousing
and Data Mining


Systems for Data Warehousing
– Arbor Software: Essbase
– Oracle (IRI): Express
– Cognos: PowerPlay
– Redbrick Systems: Redbrick Warehouse
– Microstrategy: DSS/Server
Systems or Research Prototypes for Data Mining
– IBM: QUEST (Intelligent Miner)
– Silicon Graphics: MineSet
– Integral Solutions Ltd.: Clementine
– Information Discovery Inc.: Data Mining Suite
– SFU (DBTech): DBMiner, GeoMiner
– Rutger: DataMine, GMD: Explora, U Munich: VisDB
Conclusions





Data warehousing and data mining:
– A rich, promising, young field with broad applications
and many challenging research issues.
– Imminent task: spatial database analysis --- from spatial
data manipulation to on-line spatial analytical processing
(Spatial OLAP) and spatial data mining.
Spatial data cube construction: fine granularity analysis.
Multiple spatial data mining tasks: Characterization,
association, classification, clustering, sequence and pattern
analysis, prediction, etc.
Integration of data mining with OLAP: OLAP-based spatial
data mining.
Integration of spatial analysis methods, spatial query
processing methods, and spatial indexing techniques.
Future Research





Foundation of spatial data warehousing and data mining.
Implementation methods:
– Efficient construction of spatial data cubes.
– A set of well-tuned spatial data mining operators.
– Spatial data and knowledge visualization tools.
– Integration of multiple mining tasks with OLAP functions.
New spatial indexing techniques for spatial data warehousing
and spatial mining.
New spatial data mining methodologies: Statistical tools,
neural nets, and ad-hoc query-based mining, etc.
Mining spatiotemporal data, raster data, and integration
with existing spatial analysis techniques.
References


[1] Floris Geerts, Sofie Haesevoets and Bart Kuijpers.
A Theory of Spatio-Temporal Database. Computer Science Dept., North Dakota State University (2000)


[2] Martin Ester, Hans-Peter Kriegel, Jörg Sander.Algorithms and Applications for Spatial Data Mining ,
Geographic Data Mining and Knowledge Discovery, 2001.


[3] Martin Ester, Alexander Frommelt, Hans-Peter Kriegel, Jörg Sander. Algorithms for Characterization and
Trend Detection in Spatial Databases, International Conference on Knowledge Discovery and Data Mining
(KDD-98)


[4] Jan Paredaens, Bart Kuijpers. Data Models and Query Languages for Spatial Databases. ACM SIGKDD
Explorations (1999)


[5] Hans-Peter Kriegel, Thomas Brinkhoff, Ralf Schneider. Efficient Spatial Query Processing in Geographic
Database Systems. VLDB (2001)


[6] Usama Fayyad, Gregory Piatetsky-Shapiro, and Padhraic Smyth. From Data Mining to Knowledge
Discovery in Databases. AI MAGAZINE (1999)


[7] Ramakrishnan Srikant, Rakesh Agrawal. Mining Quantitative Association Rules in Large Relational Tables.
VLDB (1996)


[8] Krzysztof Koperski, A Progressive Refinement Approach to Spatial Data Mining. SFU PhD Thesis (1999)
Thank you !!!