This document discusses building a predictive system using machine learning. It describes predicting income using census data with four machine learning algorithms: Two-Class Decision Jungle, Two-Class Averaged Perceptron, Two-Class Bayes Point Machine, and Two-Class Locally-Deep Support Vector Machine. It also discusses tuning hyperparameters, combining results, and benchmarking performance. Additional sections cover predictive analytics processes, digital transformation, and predictive maintenance maturity models.
Report
Share
Report
Share
1 of 20
More Related Content
AI Class Topic 3: Building Machine Learning Predictive Systems (Predictive Maintenance)
1. WELCOME TO:
How To Use Machine Learning To Build A Predictive
System
By Adj Prof. Giuseppe Mascarella
giuseppe@valueamplify.com
4. Target (Label): Predict Income of a person
Dataset: From Census Data
Algo:
1. Two-Class Decision Jungle
2. Two-Class Averaged Perceptron
3. Two-Class Bayes Point Machine
4. Two-Class Locally-Deep Support Vector Machine
ML Experiment
5. WELCOME TO:
How To Use Machine Learning To Build A Predictive
System
By Adj Prof. Giuseppe Mascarella
giuseppe@valueamplify.com
2. Digital Transformation
7. Generally, learning the optimal hyperparameters for a given machine
learning model requires considerable experimentation. This module
supports both the initial tuning process, and cross-validation to test
model accuracy:
• Find optimal model parameters using a parameter sweep
• Perform cross-validation during a parameter sweep
Experiment:
https://gallery.azure.ai/Experiment/b2bfde196e604c0aa2f7cba916fc
45c8
How to Configure and Tune Models
Hyperparameters
9. R Script
Combine results of all ML learners into a single column and to
add a column with the names of the algorithms.
1. dataset <- maml.mapInputPort(1)
2. .
3. Algorithm <- c("Averaged Perceptron","Bayes Point
Machine", "Decision Jungle", "Locally-Deep SVM")
4. data.set <- cbind(Algorithm, dataset)
5. .
6. maml.mapOutputPort("data.set")
Benchmark Performance
10. Value Creation | Evolving Human + Machine Intelligence
Improve Asset
Utilization ($)
Reduce Downtime($)43
56
Q1 Median
Operational
Efficiency (%)
233
301
Q1 Median
785
649
Q1 Median
f(x)
Helping you Grow, Improve Operations and Future Ready
MACHINEHuman
Market Responsiveness
(Improve Forecasting, Extend Value Chain)
Talented
workforce
Business
ecosystems
Data
access
Data and
tech
ecosystems
Agile forms of working
Machine
intelligence
Adaptive
organizations
Al-driven
Operation Excellence Future Cash Flow
(Fleet Maintainence, Asset Optimization)
Future Ready
(Sustainable Advantage)
Integrated System of
Intelligence
<->
Gaining Advantage with Intelligence & Value Creation
(Human + Machine)
<->
12. AppPlat Portfolio | Customer Centric Framework to Drive Digital Transformation
13. Digital Transformation
Microsoft, Google are focused on empowering every individual and every organization to
achieve more through Digital Transformation
Building better, stronger
engagements by harnessing
data representing a complete
view of your customer, then
drawing actionable
intelligence, predictive
insights that can deliver
personalization at scale
Engage your
customers
Reinventing products,
services and business models
using digital content to
capitalize on emerging
revenue opportunities
Transform your
products
Optimizing operations to
reshape customer
relationships and service
models by gathering data
across a wide, dispersed set
of endpoints, and drawing
insights through advanced
analytics that can be used to
introduce improvements on a
continuous, real
time basis
Optimize your
operations
Empowering employees with
tools that fuel collaboration
and productivity, while
mitigating the risks that come
with providing freedom and
space to employees
Empower your
employees
14. Analytics ValueData and analytics underpin six disruptive models and certain characteristics make individual domains
susceptible
Indicators of potential for disruption
• Assets are un-utilized due to inefficient
signaling
• Supply/demand mismatch
• Dependence on large amounts of
personalized data
• Data is siloed or fragmented
• Large value in combining data from
multiple sources
• R&D is core to the business model
• Decision making is subject to human
biases
• Speed of decision making limited by
human constraints
• Large value associated with improving
accuracy of prediction
Domains that could be disrupted
Insurance | Health care | Human capital/talent
Transportation and logistics | Automotive |
Smart cities and infrastructure
Health care | Retail | Media | Education
Banking | Insurance | Public Sector |
Human capital/talent
Life sciences and pharmaceuticals |
Material Sciences | Technology
Archetype of disruption
Business models enabled
by orthogonal data
Hyperscale, real-time
matching
Radical personalization
Massive data integration
capabilities
Data-driven discovery
Enhanced decision
making
Smart Cities | Health care | Insurance |
Human capital/intent
15. Leading Digital Transformation Through Proven Customer Use
Cases
Improving visibility
and making accurate
predictions
Getting the right
products to the
right places
Offering customers
exactly what they want,
when they want it
Fixing problems
proactively before
they start
Exploring
new business
opportunities
Remote monitoring Inventory management
Demand forecasting
Supply chain
optimization
Risk and compliance
management
Marketing mix
optimization
Personalized offers
Product recommendations
New product introduction
Predictive maintenance
Operational efficiency
Customer service
improvement
Cross-sell and upsell
Product-as-a-service
New data-driven
services
16. WELCOME TO:
How To Use Machine Learning To Build A Predictive
System
By Adj Prof. Giuseppe Mascarella
giuseppe@valueamplify.com
3. CMM (Costumer Maturity Model)
17. Predictive Maintenance | As The Capabilities To Know Your Asset Increase, Costs Decreases and OEE PerformancesTake Off
Reactive / Informative Predictive Transformative
Asset Utilization
and
Maintenance • Avoid unexpected downtime
• Avoid over- and under- maintenance
• Operate Asset as a Service
• Take preemptive corrective actions
Overall
Equipment
Effectiveness
(OEE)
• Create New Business Models
Cost of
Incidents and
Maintenance
• Free up Working Capital, Margin Contribution
• Provide Global Support
Intelligence
Reports
ERP, Maint. Data
• Value Creation through
Human and Machine Intelligence
18. CMM(Customer Maturity Model) of Predictive Maintenance | Capability Model
Predictive
Maintenance
Stage 1:
Reactive (Report)
Stage 2:
Insights
Stage 3:
Predictive/ ML
Stage 4: Transformative Stage 5:
Game Changer
OUTCOMES
Vision
Schedule and manage
using past operational and
routine performance data
Analyze conditions and
make informed decisions
Discover new insight, and
predict likelihood and
timeframe of failures
Transform the experience
with real-time insight,
actions and continuous
feedback
Shape new business
models with digital
ecosystem
Strategic
Intent
• Define operational rhythm
• Meet SLAs, compliance
and warranty conditions
• Orchestrate and leverage
readily available reports
and operational
observations
• Become purpose-driven
with connected, complete,
correct and connected
data
• Model asset-specific plans
based on the asset
condition
• Easy access to insights on
the whys and the trends
• Manage the Voice of the
Asset
• Instrument the assets to
provide real-time data on
factors affecting asset
condition
• Predict and schedule
maintenance for desired
operations
• Operate Asset as a Service
by altering the asset
behavior in real-time
• Take corrective actions
before a potential failure
• Predict and perform
maintenance based on the
business impact
• Launch digital services,
leveraging design, data
and delivery insight
• Create new customer
experiences and solutions,
integrating partner assets
• Monetize learning
KPIs
• Unplanned downtime
• Regulatory compliance
• Maintenance time and
costs
• Time between failures
• Spare parts inventory
• Annual budget
• Asset utilization
• Unexpected breakdowns
• Capital and resource
investment
• Global reach
• Revenue or throughput
per asset
• Customer loyalty
• Outcome-based pricing
• New markets
• Cross-selling
• Eco-system maturity
CAPABILITIES: Data, Intelligence and Actions
APPROACH: Architecture Directions
19. Predictive Maintenance | Capability Profile Across Maturity Levels
Predictive
Maintenance
Stage 1:
Reactive
Stage 2:
Informative
Stage 3:
Predictive
Stage 4: Transformative Stage 5:
Game Changer
CAPABILITIES
Data
(Sources, time,
quality, access)
• Manufacturers reports
• Asset features
• Failures/repairs reports
• Historical data from
operational systems
• Intermittent updates
• Asset condition data
• Correlated quality, ERP, and
operational data
• Scheduled data queries and
data polling
• Real-time, streaming data
about asset conditions,
environmental factors, and
operating conditions
• Multisite data aggregation
• Data readiness for data science
• Cognitive and feedback data
• Business process / workflow
• Organization data e.g.
operator’s skills
• Events, Smart sensing
• Ecosystem data and services
• External context (customer,
consumer)
• Real-time capability and data
discovery
Intelligence
(Interpretations,
analytics,
insights,
learnings)
• Web-based reports,
dashboards
• Data visualization of historical
and operational data
• Self-service analytics
• Asset condition monitoring
and assessment
• Statistical modeling
• Trend analysis and forecasting
• Predictions using data mining,
modeling and algorithms
across all data
• Stream analytics
• Rolling aggregates, analysis
and recommendations
• Insight at sensor and interface
levels
• Deep learning e.g. vibrations
• Real-time predictions using
current business context and
operating conditions
• Analyze current state behavior
across ecosystem and identify
opportunities
• Evaluate health of data and
algorithms and predict
adjustments
Actions
(New or change
in activities)
• Inventory assets
• Develop plans and schedule
maintenance for assets based
on past performance
• Plan and schedule resources
• Forecast and optimize
schedule and inventory
• Manage critical assets and
business operations
• Manage planned downtime
• Manage resource productivity
• Create knowledgebase
• Check health while in use
• Identify potential causes and
time window, and take
proactive actions
• Generate alerts and propose
best actions
• Support remotely
• Reliability engineering
• Self-identify alternate paths for
continuous operations
• Heal the asset while in use
• Create outcome-based
business processes and
customer experience
• Make every interaction a
source of revenue
• Productize data, intelligence,
algorithms, and business
processes
• Integrate partner services
• Create BOTs
APPROACH
Architecture
Directions
• Systems of Records
• Client/server or distributed
architecture
• Data marts
• Reporting and analytics
• Systems of Engagement
• Service-oriented architecture
• Integration
• Data warehouses
• Analytical modeling
• Systems of Inference
• Lambda architecture
• NoSQL
• Data lakes
• Cloud
• Systems of Learning
• Neural network and FOG
architecture
• Cognitive services
• In memory, edge analytics
• Systems of Digital Markets
• Microservices architecture
• APIs
20. WELCOME TO:
How To Use Machine Learning To Build
A Predictive System
By Adj Prof. Giuseppe Mascarella
giuseppe@valueamplify.com