Acides Gras insaturés Professeur Jean-Luc Olivier > Faculté de Médecine [email protected].
Data Mining - Presentation - univ-angers.frricher/dm/data_mining_2_presentation.pdf · Data Mining...
Transcript of Data Mining - Presentation - univ-angers.frricher/dm/data_mining_2_presentation.pdf · Data Mining...
Data Mining - Presentation
Dr. Jean-Michel RICHER
Dr. Jean-Michel RICHER Data Mining - Presentation 1 / 1
Outline
Dr. Jean-Michel RICHER Data Mining - Presentation 2 / 1
1. Introduction
Dr. Jean-Michel RICHER Data Mining - Presentation 3 / 1
Introduction
Flood of dataWith the flood of data available nowdays, companies areturning to analytics solutions to extract meaningful andvaluable information in order to help improve decisionmaking
Analysis and analystsThose companies need capabilities (tools, human re-sources)
to analyze historical dataforecast what might happen in the future
Dr. Jean-Michel RICHER Data Mining - Presentation 4 / 1
Introduction
Flood of dataWith the flood of data available nowdays, companies areturning to analytics solutions to extract meaningful andvaluable information in order to help improve decisionmaking
Analysis and analystsThose companies need capabilities (tools, human re-sources)
to analyze historical dataforecast what might happen in the future
Dr. Jean-Michel RICHER Data Mining - Presentation 4 / 1
Introduction
AnalystsThe people who are doing this job are called mathemati-cians, statisticians, business analysts or data scientists
Research areaExpert System (’70)Knownledge Discovery in Database (’80)Machine Learning, Adaptative LearningBusiness Intelligence, Decision MakingData Mining, Big Data
Dr. Jean-Michel RICHER Data Mining - Presentation 5 / 1
Introduction
AnalystsThe people who are doing this job are called mathemati-cians, statisticians, business analysts or data scientists
Research areaExpert System (’70)Knownledge Discovery in Database (’80)Machine Learning, Adaptative LearningBusiness Intelligence, Decision MakingData Mining, Big Data
Dr. Jean-Michel RICHER Data Mining - Presentation 5 / 1
A word on analytics
Analysis vs Analytics (Wikipedia)Analytics is the discovery, interpretation, and communica-tion of meaningful patterns in data.It generally refers to the methodology
Analytics is multidisciplinaryextensive use of mathematics and statisticsanalyze massive, complex data setsuse of software to store / collect, organize, select,process data
Dr. Jean-Michel RICHER Data Mining - Presentation 6 / 1
The Walmart 2004 example (1/4)
Walmart Databasein 2004 a series of hurricanes crossed the state ofFloridaafter the first hurricane : what customers reallywanted to buy prior to the arrival of a hurricane ?
in Walmart Retail Transaction Database one particularitem that increased in sales by a factor of 7 overnormal shopping days was foundcan you guess what it was ?
Dr. Jean-Michel RICHER Data Mining - Presentation 7 / 1
The Walmart 2004 example (2/4)
Why would people buy strawberry pop tarts ?
Dr. Jean-Michel RICHER Data Mining - Presentation 8 / 1
The Walmart 2004 example (3/4)
Why do people buy this articledo not require refrigerationdo not need to be cookedcome in individually wrapped portionshave a long shelf lifeare a snack / breakfast foodeverybody love them
Dr. Jean-Michel RICHER Data Mining - Presentation 9 / 1
The Walmart 2004 example (4/4)
a win-win partnershipWalmart stocked their stores with tons of strawberrypop tarts prior to the next hurricanessold them outWalmart wins by making the sellcustomers win by getting the product that they mostwant
Dr. Jean-Michel RICHER Data Mining - Presentation 10 / 1
2. Analytics : what, why,future, action
Dr. Jean-Michel RICHER Data Mining - Presentation 11 / 1
Kinds of analysis (1/2)
What, Why, Future and ActionDescriptive What has happened ? insight into thepast or current stateDiagnostic Why something happened ?
Predictive What could happen ? predict the futurePrescriptive Actions required to influence a particularoutcome
Descriptive, Predictive and Prescriptive analytics, are inter-related solutions helping companies make the most out ofthe big data that they have
Dr. Jean-Michel RICHER Data Mining - Presentation 12 / 1
Kinds of analysis (2/2)
What, Why, Future and ActionDescriptive: profit per store, per regionDiagnostic: why did sales go down in particular regionPredictive: which products are likely to perform betterin next quarter based on past dataPrescriptive: which customer segment shall betargeted next quarter to improve profitability
My visionpast: Descriptive and Diagnosticfuture: Predictive and Prescriptive
Dr. Jean-Michel RICHER Data Mining - Presentation 13 / 1
Kinds of analysis (2/2)
What, Why, Future and ActionDescriptive: profit per store, per regionDiagnostic: why did sales go down in particular regionPredictive: which products are likely to perform betterin next quarter based on past dataPrescriptive: which customer segment shall betargeted next quarter to improve profitability
My visionpast: Descriptive and Diagnosticfuture: Predictive and Prescriptive
Dr. Jean-Michel RICHER Data Mining - Presentation 13 / 1
The four kinds of analysis
Dr. Jean-Michel RICHER Data Mining - Presentation 14 / 1
Descriptive analysis
Descriptive Analysis - BI and DMdescribe or summarize datamake it something that is interpretable by humansnearly 80% of the data mining workloadexample: reports that provide historical insightsregarding the company’s production, financials,operations, sales, finance, inventory and customers(Dash board)is it reporting ?
Dr. Jean-Michel RICHER Data Mining - Presentation 15 / 1
Predictive analysis (1/2)
Predictive Analysis - Forecastingforecast (estimate) the future (not 100% true)set realistic goals for the business, effective planningand restraining expectationsuse various statistical and machine learningalgorithmstake existing data and fill in the missing data with bestpossible guesses
Dr. Jean-Michel RICHER Data Mining - Presentation 16 / 1
Predictive analysis (2/2)
Can be further categorized asforecasting (what if the trends continue)predictive modelling (what will happen if ...)root cause analysis (why it happens)Monte-Carlo simulation (what could happen)Pattern identification and alerts (when to correct aprocess)
Dr. Jean-Michel RICHER Data Mining - Presentation 17 / 1
Prescriptive analysis (1/2)
Prescriptive AnalysisAdvanced analysis based on
with too many choices, which one is the best ?stochastic optimizations to identify data uncertaintiesa combination of data, mathematical models andvarious business rules
Dr. Jean-Michel RICHER Data Mining - Presentation 18 / 1
3. Broaden your knowledge
Dr. Jean-Michel RICHER Data Mining - Presentation 19 / 1
The five kinds of analytics
Dr. Jean-Michel RICHER Data Mining - Presentation 20 / 1
Tools and analysis
A good combinationeach type of analytics (analysis) has its methodsanalysts buy tools to perform the analytics in discretestepsbut many decisions require a combination of dataanalysis and human experience
Dr. Jean-Michel RICHER Data Mining - Presentation 21 / 1
Use graphics
Dr. Jean-Michel RICHER Data Mining - Presentation 22 / 1
Think of what to do
Exercise: Products saleImagine you are a supermarket and you want to sell /promote five different productswhat would you do ?how would you measure if your strategy worked ?
Dr. Jean-Michel RICHER Data Mining - Presentation 23 / 1
4. Skills requirements
Dr. Jean-Michel RICHER Data Mining - Presentation 24 / 1
Skills
Skillsbe able to define the different kinds of analysis ofdata miningbe able to explain a particular analysis (like thedescriptive or predictive analysis)
Dr. Jean-Michel RICHER Data Mining - Presentation 25 / 1
4. End
Dr. Jean-Michel RICHER Data Mining - Presentation 26 / 1
University of Angers - Faculty of Sciences
UA - Angers2 Boulevard Lavoisier 49045 AngersCedex 01Tel: (+33) (0)2-41-73-50-72
Dr. Jean-Michel RICHER Data Mining - Presentation 27 / 1
Dr. Jean-Michel RICHER Data Mining - Presentation 28 / 1