No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

No BI without Machine Learning

Francis Pierautfrancis@qmining.com

http://fraka6.blogspot.com/

10 March 2011MTI-820 ETS

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

To Much Data

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Why a talk about machine learning and BI?

Machine Learning 101Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

Let’s dive into practical exampleTarget MarketingCustomer behaviorRetentionRisk AnalysisMonitoring Root Cause Analysis - QMonitorMonitoring Root Cause Analysis - QMiner

Conclusion

Questions?

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Why a talk about machine learning and BI?

I Machine Learning ⇒ Data-Mining ⇒ BI

I Prediction or Clutering ⇒ Patterns ⇒ Patterns (revenus $$ ⇑)

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Speaker: Francis Pieraut, P.Eng. M.Sc.A.

I Master@LISA - Statistical Machine Learning - udm(flayers: C++ Neural Networks lib)

I Industry - 7 years in Machine Learning/AI startups

(mlboost: Python Machine Learning Boost lib)

I Founder QMining

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

AI and Machine Learning - Data-mining

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

Machine Learning and Data-Mining

I Machine Learning: learn from data

I Data-mining: extracting patterns from data

I Machine Learning use extracted patterns to do prediction

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

Machine Learning

I Learning from data

I Classification vs Clustering

I Applications: Attrition, Rank Customer (approve loans andcredit card),Fraud detection, Target-Marketing, Risk Analysis(insurance) etc.

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

Supervised Learning (need class tag for each example)

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

Unsupervised Learning - dimension reduction/clustering

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

Learning Process

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

Tanks in the desert (black box danger)

I Using ML requires insights

I An algo is only goods as its data

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Supervised Learning (classification)Unsupervised Learning (clustering)Training and TestingImportant Concepts

Important Concepts

I Datasets (features + class)I Generalization vs OverfittingI Classification vs ClusteringI Features Quality (invariant and informative)

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Target MarketingCustomer behaviorRetentionRisk AnalysisMonitoring Root Cause Analysis - QMonitorMonitoring Root Cause Analysis - QMiner

Service provider

I Find most probable interested clientsI N most likely to buy (sort DESC probability)I google mail

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Target MarketingCustomer behaviorRetentionRisk AnalysisMonitoring Root Cause Analysis - QMonitorMonitoring Root Cause Analysis - QMiner

Cell phone usage

Find users cluster (behavior)

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Target MarketingCustomer behaviorRetentionRisk AnalysisMonitoring Root Cause Analysis - QMonitorMonitoring Root Cause Analysis - QMiner

Service provider

I Find most probable clients to quit

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Target MarketingCustomer behaviorRetentionRisk AnalysisMonitoring Root Cause Analysis - QMonitorMonitoring Root Cause Analysis - QMiner

Insurance

I Score customer risk of making a claim

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Target MarketingCustomer behaviorRetentionRisk AnalysisMonitoring Root Cause Analysis - QMonitorMonitoring Root Cause Analysis - QMiner

QMonitor - Global Server Incident Mining

I Find incidents on serversI Find patterns (network, server, etc.)

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Target MarketingCustomer behaviorRetentionRisk AnalysisMonitoring Root Cause Analysis - QMonitorMonitoring Root Cause Analysis - QMiner

QMiner - Global User Experience Incident Mining

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

What you should remember?

I No BI without Machine Learning

I Machine learning algorithms applications ⇑I goal = generalization⇒good prediction (DON’T OVERFIT)

I 80-90% pre or post-processing + data visualization

I Python provide amazing integration

I **QMining is looking for intership students

I BI for Business User http://www.qlikview.com/

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning

OutlineWhy a talk about machine learning and BI?

Machine Learning 101Let’s dive into practical example

ConclusionQuestions?

Any questions?...intership 2011 ⇒ francis@qmining.comhttp://fraka6.blogspot.com/..Thanks,Francis Pierautfrancis@qmining.com

Francis Pieraut francis@qmining.com http://fraka6.blogspot.com/No BI without Machine Learning