Data Mining
www.keysoft.co.in
Copyright KEYSOFT Solutions
Agenda What What is Data ata Minin ining? g? Definition CRISP-DM Mode Modeli ling ng Techn echniq ique ues s How How mod model els s are are bui built lt and and use used? d? Data Data ware wareho hous use e Vs Data Data minin mining g Appl Applic icat atio ions ns Of Of Dat Data a Mini Mining ng Case Study - Telec lecom
www.keysoft.co.in
Copyright KEYSOFT Solutions
What
is Data Mining?
Inte Intera ract ctiv ive e and and itera iterati tive ve pro proce cess ss To find under underly lying ing relat relation ionshi ships ps and and feature features s in data data Knowledge dr driven Enhanc Enhance e the the valu value e of exist existing ing info informa rmatio tion n reso resourc urces es Pred Predic icti ting ng valu valuab able le inf infor orma mati tion on Data Data or kno know wledg ledge e dis disco cov very ery To predi predict ct future future trend trend and behav behavior ior
www.keysoft.co.in
Copyright KEYSOFT Solutions
Definition The process of finding previously unknown patterns and trends in databases and using that information i nformation to build predictive models.
www.keysoft.co.in
Copyright KEYSOFT Solutions
CRISP
- DM
CRIS RISP-D P-DM stands ands for for Process
CRoss Industry Standard
for Data Mining
It is is a indust industry ry and and tool tool neut neutral ral data data mining mining proces process s model This This Proces Process s makes makes large large data data minin mining g project projects s faster faster,, cheaper and more reliable
Check CRIP-DM¶s home page for more details www.crisp-dm.org
www.keysoft.co.in
Copyright KEYSOFT Solutions
Phases
www.keysoft.co.in
in
CRISP
- DM
Copyright KEYSOFT Solutions
Modeling Techniques Predictive
Models
Neural Nets Rule Induction Line Linear ar and and Log Logis isti tic c Reg Regre ress ssio ion n Example
Credit card Company can rate the customer before issuing Credit card by using the predictive models. Customer¶s applications which are rated low by the pred predic icti tiv ve mode models ls are are reje reject cted ed..
www.keysoft.co.in
Copyright KEYSOFT Solutions
Modeling Techniques Association Rules
APPRIORI GRI Example
A leading Super market applies association techniques in its transaction database and finds out items which are often purchased together and comes up with ith new bund bundle led d offe offerr to prom promot ote e its its othe otherr non non selli elling ng item items s.
www.keysoft.co.in
Copyright KEYSOFT Solutions
Modeling Techniques Clustering
Techniques
Kohonen Ne Networks K-Means Two Step Example
A Insurance company groups similar customers based on various parameters and launches special promotional offers targeting each segment and achieves high higher er resp respon onse se rate rate..
www.keysoft.co.in
Copyright KEYSOFT Solutions
How
Models are built and used?
www.keysoft.co.in
Copyright KEYSOFT Solutions
Data Mining
www.keysoft.co.in
Process
Copyright KEYSOFT Solutions
Data Warehouse Vs Data Mining Data Warehouse
Data Mining
Contains historical Process of finding hidden information stored stored in the information from large form of relational database datasets. which are acquired from different sources. Present information can be acquired.
Future information can be predicted.
Provides answers to questions like
Provides answers to questions like
³Who is purchasing our products?´
³Who is not purchasing our products?´
www.keysoft.co.in
Copyright KEYSOFT Solutions
Applications
of Data Mining
Marke arkett Seg Segme ment ntat atio ion n Customer ch churn Fraud Detection Direct Ma Marketing Inte Intera ract ctiv ive e Mark Market etin ing g Mark Market et bask basket et Anal Analy ysis sis Trend An Analysis
www.keysoft.co.in
Copyright KEYSOFT Solutions
tudy y Case Stud
- Tele eleco com m
Objective:
To retain the customer base for a telecom company by creating a model using Neural network algorithm which will pred predic ictt which hich cust custom omer ers s are are prob probab able le chur churne ners rs
www.keysoft.co.in
Copyright KEYSOFT Solutions
tudy y Case Stud
- Tele eleco com m
Data Preparation
Use the serv service ice usage usage data data whic which h contai contains ns summ summari arize zed d information about present as well as past customers Prepar Prepare e the the data data by remov removing ing miss missing ing value values s and deriv deriving ing new fields Partit Partition ion the the data data as Trai Trainin ning g and Testing esting data data Select Select only only importa important nt field fields s as input input field fields s with with respec respectt to output field
www.keysoft.co.in
Copyright KEYSOFT Solutions
tudy y Case Stud
- Tele eleco com m
Modeling
Apply Apply Neura Neurall Network Network Algori Algorithm thm to Trai Trainin ning g data data and build build the model After creating creating the the model model apply apply Testing esting data data in order order to verify verify the models accuracy Deployment
Deploy Deploy the fina finall model model to the the curre current nt cust custome omerr datab database ase The mode modell Predi Predict cts s µn¶ no. of cust custome omers rs are prob probable able Churners in its current customer database
www.keysoft.co.in
Copyright KEYSOFT Solutions
tudy y Case Stud
- Tele eleco com m
End Result
The telecom company reacts immediately by sending lucrative offers through SMS and other means to these probable churners by then it prevents its customers from losi losing ng to its its comp compet etit itor ors s
www.keysoft.co.in
Copyright KEYSOFT Solutions
Finally«.
Data mining lets you be
www.keysoft.co.in
Proactive
rather than Retrospective
Copyright KEYSOFT Solutions
For
more presentations visit:
http://www.keysoft.co.in/downloads.aspx Or mail us at
[email protected] For Forum
an exclusive
on Business Intelligence visit: http://forums.keysoft.co.in
www.keysoft.co.in
Copyright KEYSOFT Solutions
