AI on Demand: Data Science in Operations

© Copyright 2000-2018 TIBCO Software Inc.
AI on Demand
Data Science in Operations
Michael O’Connell
Chief Analytics Officer
TIBCO
Twitter: @MichOConnell

2
AI Success Stories
FINANCE:
Fraud Detection
MANUFACTURING:
Anomaly Detection
HEALTHCARE:
Patient Diagnosis
TRANSPORTATION:
Demand Forecasting
E-COMMERCE:
Recommender Systems
RETAIL:
Inventory Optimization

3
Data Flow: Event — Data — Insight
1/ All Data Begin as
Real-Time Events
2/ Analytics on
Accumulated Data
3/ Insights are Perishable
=> Take Action !!

4
AI has been around - my AI timeline
1991 - 1997 2007 2008 - 2010 2012 2016 2018
KBS & Statistics Machine Learning Data Mining KDD Big Data Artificial Augmented
Ensembles Intelligence Analytics

5© Copyright 2000-2019 TIBCO Software Inc.
Recommender Systems
Suggestion Engines
E-COMMERCE:
Recommender Systems

6
Typical Applications of
Recommender Systems
Consumer Recommender Apps
• Movies (Netflix)
• Songs (Spotify)
• Books (Goodreads)
• eCommerce Products (Amazon)
AI Suggestion Engines
• Collaborative Filtering
• Factor Analysis / Latent Variables
• Content-based Filtering

7
Netflix “The Goal is to Become HBO before
they Become US”
2007 2011 2013 2018
$1M competition House of Cards Netflix on Top
Recommender Deal with NO PILOT Beat HBO
2018 : $8B on Original Programming
Most Emmy Award Noms – beat HBO
2007 : BellKor win Netflix $1M challenge
New AI Suggestion Engine (ML Ensemble)
2012 : 113 Pilots, $400M, 13 to season 2
Netflix grabbed House of Cards – NO PILOT

8
Spotfire A(X) Analytics Experience – Today!
Augmented
Search & AI-Powered Insights
Start in seconds, instant insights
Automated
Automagical Dataflows
Author & audit with automatically
recorded dataflow steps
Agile
Reimagined User Interface
Agile exploration made even easier
Accelerated
Real-time Insights
Real-time awareness and action

9
Spotfire X - AI Recommendations
Variable Relationships – Suggestion Algorithm
• User selects target variable
• AI algorithm suggests variable relationships to target
• Spotfire displays in order of strength - best practices graphics
• 3 clicks to brush-linked analytic application
Search and NLQ
• User searches multiple data tables + relationships
• Spotfire displays appropriate best-practice graphics
• User asks Spotfire how to do software functions
• Brush-linked dashboard entirely from chat interaction
Auto insight discovery & display

10
Spotfire AI Recommendations: Telco Churn
1. Target = Customer Status 2. AI Suggestions = Churner ‘friends’, overcharging, handset

Spotfire AI Recommendations: Energy
1. Target = Energy Production 2. AI Suggestions = GeoLocation, Geology, Sensor Data

Select a Target See Graphs
Data
preparation
Feature
relationships
Ranking
+
MORE LIKE THIS shows extra graphs with the same features
Adds graph to the page
Analyst Business User
Tell me about “Target”
Continue on the journey: adorn, manipulate, explore, drill
Spotfire Recommendations Visual Analytics

13
Anomaly Detection
MANUFACTURING:
Anomaly Detection

14
Typical Applications of Anomaly Detection
IoT & Engineering
• Formula One
• Energy - Production Surveillance, Drilling Optimization
• Predictive Maintenance (PdM, CbM)
• Manufacturing - Yield Optimization
Financial Services
• Trade Surveillance
• Fraud Detection
Healthcare & Pharmaceutical
• Patient risk – cardiac arrest, sepsis, surgery infection
Customer Analytics
• Churn Prevention
• Cross-Sell, Up-Sell
Key Issue – Understand Variability

Mercedes AMG Petronas
Motorsport
Car R&D
Setup, Configuration
Race Strategy

16
Data Science at Mercedes
Data created during a weekend is 400GB+
Drivers – Teams, Base Times
Outings – Tyres, Fuel
Tyres – Type, Condition
PitStop – Lap, Issues
Lap and Sector – Car position
GPS on all Cars – Resolution in cm
R&D – Car Design (Data Science, Streaming)
• Wind Tunnel – Optimize fixed time allocation
• Reliability e.g. Gearbox – Anomaly Detection; Optimal configuration
Car Setup (Spotfire)
• Visual Analytics on configurations and components
Race Strategy (Data Science, Spotfire)
• Overtakes – Machine Learning model
• Simulations for individual race tracks

AI on Demand: Data Science in Operations

Energy—it’s a Digital Business
Downstream
Midstream
Upstream
Services : Drilling
& Equipment
Exploration,
Development,
Production
Refining &
Supply
Performance risk &
trading
Fuels &
Marketing
Strategic Planning
Planning &
Scheduling

Production Surveillance – IIoT Core
Control
Center

Drilling Optimization – IIoT Edge

Anomaly Detection
Hi Tech Manufacturing
Fault Detection
Run to Run Control
Yield Optimization

Feedforward Run-to-Run
controller
Run-to-Run
controller
Feedback Feedback
Advanced Process Control – IoT Analytics
Shop Floor: Tools
Metrology Process ToolProcess Tool
Process
flow
Metrology
Fault Detection
models
Fault Detection
models
ACTIONS
MES
Tool Maintenance
MES
Product Hold
Email
Notification
Stop Eq.
Process
Stop
process
Fault Detection, Virtual Metrology
Run-to-Run control
Edge Tier: Pre-process
Edge analytics:
Extract useful features
Flogo
Platform Tier: Real-time Decisions
Streambase
Enterprise Tier: Visualize & Manage - Business Apps
Spotfire, Data Virtualization
Reports & Dashboards
Virtual Metrology
Data Science Tier:
TIBCO Data Science, Data Lake
Models: Build, Evaluate &
Push to Production

Real-time Factory Equipment Monitoring
• Global View
• Data from MES, Equip sensors
• Monitor Factory Metrics: OEE
• Factory View
• Monitor Equipment Status –
predictive & condition-based
maintenance
• Sensor View
• Statistical Process Control:
• AI-based anomaly detection
Hi Emissions
Cluster

Deep Learning Autoencoder for Anomaly
Detection: Spotfire & Data Science & Streaming

28
HEALTHCARE:
Patient Diagnosis
Patient Diagnosis

29
Patient Diagnosis &
Intervention
Analytics & Real-Time Surveillance
• Cardiac Arrest Monitoring (UCM)
• Surgical Site Infections (U Iowa)

30
The Surgical Site Infection Problem
• Cost of hospital acquired infections
in US is in excess of $45 billion
• Readmissions lead to $41.3 billion in
costs
• SSI #1 cause of HAI - 21.8%
• SSI #1 cause of readmission in
surgical patients - 38.8%
Magill, S.S., et al. (2014). Multistate point-prevalence survey of health care-associated infections. N Engl J Med 370, 1198–1208.
Kassin, M.T., et al. (2012). Risk Factors for 30-Day Hospital Readmission among General Surgery Patients. J Am Coll Surg 215, 322–330.

31
Outcomes — Better than a Pill
58% reduction in SSI in 2 years
74% reduction in SSI in 3 years
Estimated cost reduction during hospitalization:
$2.2 million for every 300 patients undergoing surgery
Gbegnon, A., et al. (2014). Predicting Surgical Site Infections in Real-Time. KDD Proceedings
Cromwell, J. (2018). Machine Learning and High Definition Care: More Powerful than a Pill. TIBCO NOW Proceedings, community.tibco.com

32
Data Flow
TIBCO
Data Science

33
Important Variables, Model Summary
Variable K- NMI P value
SSIs 1 < 0.001
*Duration 0.035 < 0.001
*EBL 0.043 < 0.001
*Ostomy 0.055 < 0.001
Procedure category 0.029 < 0.001
Surgeon 0.021 < 0.001
Total procedures 0.028 < 0.001
*Transfusion 0.028 < 0.001
*Wound class 0.014 < 0.001
Zip code 0.032 0.13
Location 0.013 0.11
OUTCOME
No SSI SSI
No SSI TN=88 FN=5
SSI FP=27 TP=6
* = variable collected in OR in real-time
COST
SSI $28,000
Intervention $500
Break-even 56 FP = 1 FN

34
End-State Architecture
EMR
IntegrationLayer
API
Management
Offered
API
Offered
API
Offered
API
Data
Repository
Statistical
Model
Statistical
Model
Analytics
Server
Statistical
Model
Provider Environment Service Environment
TIBCO SaaS Offering
BusinessWorks
Container
Edition
Statistica
Spotfire

In the News

36
AI for Social Good

37
Impact of parking citations on low-income
drivers in San Francisco
TIBCO DATA SCIENCE (ML on big data) with TIBCO SPOTFIRE Visual Analytics
TIBCO Products:
• TIBCO Data Science
• TIBCO Spotfire
Business objective: Discover insights and
make recommendations to SF Office of
Financial Justice for policy review
Collaboration between Tipping Point, SF
Office of Financial Justice and TIBCO
Delivers:
• Big Data – data prep and analytics at
scale
• Visualization (in-db and or in-memory)
of reduced dataset
• Simplified Interface for Business Users
• Web / cloud deployment (AWS)

44
Change in Policy!

Community and Resources
Ecosystem to help customers succeed with analytics and drive adoption
community.tibco.com
community.tibco.com

46
Questions & Contact
Thank you!
Michael O’Connell
Chief Analytics Officer
TIBCO
moconnell@tibco.com
@MichOConnell

47
AI on Demand Community Site
https://community.tibco.com/wiki/ai-demand-data-science-operations
Watch this TIBCO Community
Site for updates
• AI on Demand demos
• AI on Demand case studies
• AI on Demand tour dates

AI on Demand: Data Science in Operations

More Related Content

What's hot (19)

Similar to AI on Demand: Data Science in Operations (20)

Recently uploaded (20)

AI on Demand: Data Science in Operations