Software with RapidMiner Data Analytics @ Lufthansa Agenda 1 - - PowerPoint PPT Presentation

From Prototype to Operative Software with RapidMiner

Data Analytics @ Lufthansa

Agenda

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 2

4

Lufthansa Industry Solutions & RapidMiner

3

FlightPrediction @ Lufthansa

2

Our Daily Challenges

1

Lufthansa Industry Solutions: Who We Are

IT Consulting and Systems Integration

Lufthansa Industry Solutions combines the dynamics of an SME with the economically secure background of Lufthansa; an internationally acting global corporation.

9/1/2017 Company Presentation 3

Lufthansa Industry Solutions

We are

Services overview

9/1/2017 Company Presentation 4

Lufthansa Industry Solutions

IT Consulting and System Integration Strategy and Management Consulting IT Service Provider

Strategy Consulting IT System Operations IT System Integration & Development IT Consulting

(processes, technologies, infrastructure)

Operations of full business processes BPO specialized provider Organization & Process Consulting

Process Consulting / Process Organization Organi- sational structure Infra- structure Application

• Mgmt. &

Operations

Norderstedt

Hamburg

Cologne

Berlin

Wetzlar

Frankfurt

Basel

Miami Wolfsburg Oldenburg

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 5

Locations

Our

Norderstedt

Hamburg

Cologne

Berlin

Wetzlar

Frankfurt

Basel

Miami Wolfsburg Oldenburg

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 6

Locations

Our

Hard Facts

100% owned by

Lufthansa

208 Mil. €

total revenue (46% of which within the Lufthansa Group)

>200

customers

>1300

trained employees

Managing Director

Bernd Appel

Founding

1998 as Industry Solutions division within Lufthansa Systems AG

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 7

AirPlus LGBS Lufthansa Passenger Swiss Austrian Lufthansa Industry Solutions Lufthansa Systems

Finance and Service Companies Passenger Airlines Digital Service Companies

Lufthansa Industry Solutions

Lufthansa Technik LSG Sky Chefs

Aviation Industry

Lufthansa Cargo

Air Freight

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 8

Projects Successfully Delivered

Many

About Me

Fabian Werner Born 1985 Ph.D. Mathematics (Number theory)

Past Present Future

• Predictive analytics:
• Classification,

regression, recommender systems, probabilistic modelling, …

• RapidMiner, R, Python,

PHP, Teradata, Oracle, Hadoop, …

• Probabilistic modelling
• Time series analysis
• Bayesian learning
• Learning theory

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 9

Quick Facts

Our Daily Challenges

Lufthansa Internal

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 10

Dashboarding Predictive Analytics (Deep) Machine Learning Data Ingestion

• Airplane punctuality
• Competitor information
• Airfreight load monitor
• Arrival times
• Passenger connection
• Fuel & Weight forecast
• Predictive maintenance
• Recommend fligths
• Document and image

recognition

• Damage recognition

via audio data

• Bots (see Mildred)
• Live flight data
• Weather data streams
• Airplane sensor data

Process Model

Scoping Phase

Laboratory stage Implementation stage

Laboratory stage Implementation stage

11 9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa

FlightPrediction @ Lufthansa

We predict the arrival time

• n aircraft takeoff

The Cycle

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 13

Goal & Process

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 14

Landing Takeoff Departure Arrival

Rolling Rolling Airtime Gate - to - Gate

Prediction Point Prediction Time

Goal & Process

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 15

Landing Departure

Rolling Rolling Airtime Gate - to - Gate

Arrival Prediction Point Takeoff Prediction Time

Goal & Process

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 16

Landing Departure

Rolling Rolling Airtime Gate - to - Gate

Arrival Prediction Point Takeoff Prediction Time

Accuracy of existing prediction

• Avg. deviation estimated vs. actual OnBlock

Goal & Process

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 17

Landing Departure

Rolling Rolling Airtime Gate - to - Gate

Arrival Prediction Point

Accuracy of existing prediction

• Avg. deviation estimated vs. actual OnBlock

Takeoff Prediction Time

How? Unleashing The Power of Big Data

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 18

Predicted Arrival Time Historized Flight Data Stream Eurocontrol Weather Data Stream Data Warehouse Reporting Server Real-Time Data Historic Data

The Cycle

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 19

Parameters for the Model

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 20

Flight Duration Passenger data

• #Passengers with connection
• % occupied seats

Route, distance, … Weather

(en route & @destination)

Season Lagged features Runway(s) Weight Traffic situation Delay @ Takeoff

Parameters for the Model

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 21

Flight Duration Passenger data

• #Passengers with connection
• % occupied seats

Season Lagged features Runway(s) Weight Delay @ Takeoff Special disruptions Actual data needed Route, distance, … Weather

(en route & @destination)

Traffic situation Capacity index & airport information still missing

The Cycle

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 22

• A few rolling window features are good!
• Prefer models that can be trained fast:

Play around with features instead of models.

Model Building

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 23

The Cycle

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 24

Visualization: Performance Monitor

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 25

Data Made Up (No Actual Data)!

10 20 30 40 50 60 70 80 90 100 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

Error @ flight done %

• ther prediction
• ur prediction

10 20 30 40 50 60 100 90 80 70 60 50 40 30 20 10

Error @ minutes before arrival

• ther prediction
• ur prediction

Error Error

How To Save Money?

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 26

Gate Z50 Gate Z99

How To Save Money?

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 27

Gate Z50 Gate Z99 Gate A1

How To Save Money?

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 28

Gate Z50 Gate Z99 Gate A1

Visualization: Benefits Monitor

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 29

Data Made Up (No Actual Data)!

1 2 3 4 5 1 2 3 4 5 6 7

Potential benefit (k€)

Saved Passengers Potential Benefit

Potential Benefit 1000*€ #Saved Passengers

The Cycle

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 30

• Runs every hour
• Asserts that …
• enough data
• enough predictions
• predictions produced fast
• not too much missing data
• …

Automation: Supervisor

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 31

Supervision of Model

• Supervisor & dashboard as early as

possible

• Business needs to understand the

real impact

• Translate Data Science to business

language

• Need for standards
• Prototype needs to be translated to

real application

• You do not want to rebuild everything

Lessons Learned

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 32

Go To The Real World Fast – Make Everybody Understand The Benefits

• Partnership founded in 2015
• RM widely used in Lufthansa
• Keeps up with market technologies
• For prototyping and productive applications
• Still lacks some software engineering capabilities …
• We pair it up with Hadoop, Oracle, Teradata, R, Python, …
• Lufthansa Industry Solutions: certified RapidMiner trainers

Partnership

Lufthansa Industry Solutions | RapidMiner

9/1/2017 From Prototype to Operative - Data Analytics @ Lufthansa 33

Many Thanks for Your Attention!