LYTIQ GmbH Vorhersagen besser treffen mit Predictive Analytics Köln, April 2015 Chief Data Scientist Dieser Bericht enthält vertrauliche Informationen und ist ausschließlich für die interne Nutzung von Mitarbeitern der ORCONOMY GmbH sowie Kundenmitarbeitern bestimmt. Die Nutzung, Verteilung, Zitierung und Vervielfältigung auch auszugsweise zum Zwecke der Weitergabe an Dritte ist nur nach schriftlicher Genehmigung der ORCONOMY GmbH gestattet. Wenn Sie nicht der rechtmäßige Empfänger dieses Berichts sind, informieren Sie uns bitte über den Erhalt dieser Unterlagen und löschen diese.
Agenda Introduction Predictive Analytics Use Cases An Azure Architecture for Predictive Analytics Azure Machine Learning Demo Page 2
Introduction Leveraging Predictive Analytics and Machine Learning - From structured and unstructured data - For Big Data and small data scenarios
LYTIQ employees have worked for the following customers Page 4
What is Machine Learning / Predictive Analytics Find out how Machine Learning & Predictive Analytics works to better understand where and how to apply it.
What is Machine Learning (ML) Making Predictions Customer attributes & past transactions Purchase Indicator X Input Data y Target Data Page 6
What is Machine Learning (ML) Making Predictions Customer attributes & past transactions Purchase Predictions So we try some functions h that we do know and see which does best h( ) X y Input Data Target Data h(x) generates predictions y. Should be close to original y. Page 7
What is Predictive Analytics (PA) It is an oversimplification and as with all marketing terms there is lots of dispute But to me: Predictive Analytics is applying ML to business problems. I.e.: Customer churn should be predicted. Historic data is available. Available data actually explains that what should be predicted. Data is shaped to differentiate between customers in the various lifecycle states. Now ML models are built to score the customer s propensity to churn giving his behaviour. Page 8
Machine Learning / Predictive Analytics Use Cases Find out how Machine Learning & Predictive Analytics can help you make complex decisions better and faster.
Predictive Analytics Now Span Several Industries Telemetry Data Analysis Buyer Propensity Models Social network analysis Predictive Maintenance Web app optimization Churn Analysis Natural resource exploration Weather forecasting Healthcare outcomes Fraud detection Life sciences research Targeted Advertising Network Intrusion Detection Smart meter monitoring Page 10
Predicting Part Maintenance 1 No sensor for measuring the thickness of a brake pad or wear of a tire Need to predict when a failure will occur. 2 Inform the driver Change of parts might be necessary in a few weeks Order parts and prepare an individual maintenance package before the driver comes to the garage 3 Analytical Task: Predict the possibility of next break pad failure using data collected from sensors and telemetry
Use Case: Customer Segmentation 1 Assign drivers into categories Sports driver, Grandma / Grandpa driver 2 Provide offerings based on driver type Sports package for the sports driver Make a special offer for the next car Sporty look for the sports driver 3 Analytical Task: Cluster the relevant diagnostics data to find groups of drivers with common driving habits Page 12
Use Case: Improving Vehicle Recalls Additional Data Collection for selected vehicles Archive Aggregated Data OnBoard Storage On service contact Scored Vehicles Page 13
Use Case: ATM Failure Prediction Page 14
Other IoT Use Cases Improved Production Monitoring - Trace errors in production to machine components and process steps - Dynamic production tolerances Product Lifecycle Management - Find true requirements for new components - Improve Weibull Bayes models Page 15
Use-Case: Targeted Mailing Optimization CMS Campaign- Management DWH Page 16
Other Predictive Analytics Use Cases Customer Analytics - Unified Customer View - Customer Segmentation - Market Segments - Customer Clusters - Full individual-level personalization - Customer Retention / Churn Detection - Prospective Client Analysis - Dynamic Personalization / Recommendations - Cross-Selling, Up-Selling - During car configuration - For direct sales contact Page 17
How about Elevators & Escalators? Data from *lots* of machines having dozens of sensors Data generated per minute (practically up to seconds) Event Hubs Complex Event Processing Page 18
Predicting & Optimizing Traffic Full routes known Predicting ETA / trip time Improving ETA predictions Predicting load on route segments Detect traffic jams and consequences Central traffic planning? No, social gaming Load based traffic pricing and taxing? Improved London Congestion Charge Page 19
Improve Logistics to Reduce Traffic Logistics Marketplace Project proposed (BMBF) Central logistics planning DB & service Increase transparency Reduce traffic Page 20
Ground Image Data Commercial satellite imagery available on at least daily basis planet.com, 2x per day spire.com, 24x per day 5m or better GSD Improve traffic models! Detect roads, (illegal) construction, land use, Can tell different kinds of trees or crops apart Page 21
Other Topics Joint intelligent transportation systems Trains, long distance & local commute Rental cars Cabs Bicycles Electricity Water Waste Pollution Emergency prediction & management Other public services Page 22
Microsoft & Machine Learning Microsoft & Machine Learning 20 years of realizing innovation 1999 2004 2005 2008 2010 2012 2014 Computers work on users behalf, filtering junk email Microsoft search engine built with machine learning SQL Server enables data mining of databases Bing Maps ships with ML trafficprediction service Microsoft Kinect can watch users gestures Successful, real-time, speech-tospeech translation Microsoft launches Azure Machine Learning, making years of innovation available John Platt, Distinguished scientist at Microsoft Research Machine learning is pervasive throughout Microsoft products. Page 23
What is Azure ML Page 24
Advanced analytics architecture API HDFS RDBMS NoSQL stores Blobs and tables ML Studio Marketplace Page 25
Revolution Analytics R Acquisition still pending close. Treat them as a partner until then. http://blog.revolutionanalytics.com/2015/01/revolution-acquired.html http://blogs.microsoft.com/blog/2015/01/23/microsoft-acquire-revolutionanalytics-help-customers-find-big-data-value-advanced-statistical-analysis
Microsoft and Revolution Analytics Strategic Rationale
Revolution R Open Revolution R Enterprise R Language Engine with multi-core processing Included Supported R Reproducibility Toolkit & MRAN Included Supported ParallelR: Parallel Programming Toolkit RHadoop: R interface to Hadoop MapReduce DeployR Open: Web Services API RRE DeployR Multi-server, enterprise authentication RRE ScaleR Big Data toolkit and PEMAs for R RRE DistributedR EDW, Grids, Hadoop AdviseR Technical Support Open Source Assurance Supported Supported Supported Licensed & Supported Licensed & Supported Licensed & Supported Included Included Revolution Analytics Services (Consulting / Training) Available Available 28
Churn Prediction // CH_M5 We reduced our customer turnover by 25% using this model, it identified customers most at risk and enabled us to improve our relationship overall with our customers. Alice Wing, Customer Relations Director, Contoso Corp Search for a model All Industries Banking Education Page 29
LYTIQ GmbH is a Solution Provider for Predictive Analytics. Whether Big Data or Small Data - analytics we are your reliable partner. Contact person: Jun.-Prof. Dr. Krohn-Grimberghe, Artus LYTIQ GmbH Technologiepark 19 33100 Paderborn +49 5251 682207-0 artus@lytiq.de www.lytiq.de Page 30