Predictive Analytics World Berlin
13.-14. November 2017, Estrel Hotel Berlin

Erste bestätigte Keynotes und Sessions

Das komplette Programm und die genaue zeitliche Abfolge der einzelnen Sessions wird noch bekannt gegeben.


How Predictive Modelers Should use Data to Tell Data Stories

As data science captures more attention from decision makers, translating the models from the language of the analyst into a language of the decision maker has become an important topic at conferences and in journals. It used to be that the focus on data storytelling was on visualization techniques. While this is important, as analyses become more complex, the task of interpreting the models likewise becomes more complex. Before we can decide on visualization techniques, we first need to uncover what to visualize. In this keynote, Mr. Abbott will describe ways to unravel complex descriptive and predictive models so they can be explained and visualized using machine learning models and resampling techniques.

Dean Abbott, Co-Founder and Chief Data Scientist, SmarterHQ


Data Alchemy

The “Big Data” and “Data Science” rhetoric of recent years seems to focus mostly on collecting, storing and analysing existing data. Data which many seem to think they have “too much of” already. However, the greatest discoveries in both science and business rarely come from analysing things that are already there. True innovation starts with asking Big Questions. Only then does it become apparent which data is needed to find the answers we seek.
In this session, we relive the true story of an epic voyage in search of data. A quest for knowledge that will take us around the globe and into the solar system. Along the way, we attempt to transmute lead into gold, use machine learning to optimise email marketing campaigns, experiment with sauerkraut, investigate a novel “Data Scientific” method for sentiment analysis, and discover a new continent.
This ancient adventure brings new perspectives on the Big Data and Data Science challenges we face today. Come and see how learning from the past can help you solve the problems of the future.

Lukas Vermeer, Data Scientist,

Deep Dive (Beginners):

Time series: Dynamic Forecasts with Bayesian Models and Neural Networks

Time series data are everywhere: In finance, accounting, sales, IoT (sensor data), operations, production, marketing, economy, ecology … even in our daily lives, where we might measure series of daily calory intake, daily energy consumed, work productivity over the day… and what not. Given the past data, how can we forecast the future? In this session, after a quick look at the classics (random walks, ARMA and ARIMA models) we’ll dive into two lesser known but extremely interesting alternatives: Bayesian dynamic linear models (including the famous Kalman filter) and the deep learning approach to sequence modeling, recurrent neural networks (mainly the LSTM). All demos will be available to the participants as R notebooks to play with.

Sigrid Keydana, Data Scientist

Deep Dive (Expert):

Deep Recurrent Neural Networks: Theory and Industrial Applications

The talk focuses on a new hype wave in neural network modeling called Deep Learning or Deep Neural Networks. We will look behind the scenes and will explain the differences between “standard” feedforward and deep neural network models. The decay in gradient information over long chains of hidden layers can be avoided by e.g. multiple outputs, information highways or short cuts, as well as stochastic learning rules. Auto-associators and convolutions enable us to model high-dimensional input spaces by the automated generation of features. Besides deep feedfoward neural networks we will also deal with time-delay recurrent neural network architectures, where deepness is a natural feature when non-algorithmic learning techniques like error backpropagation through time are used. We will give examples of the application of deep neural networks form recent project work and show the merits of the “new” approach in comparison to non-deep modeling techniques. Among others we will deal with the modeling of e.g. the energy supply from renewable sources, energy load forecasting as well as the identification of features responsible for component failures and soft sensor modeling.

Dr. Ralph Grothmann, Principal Consultant, Siemens AG, Corporate Technology

Case Study:

Der Wert von externen Daten für Predictive Analytics im B2B-Umfeld am Beispiel von Swiss Re

Die Swiss Re als Rückversicherer operiert – auch im Bereich Predictive Analytics – als B2B-Geschäftspartner für Erstversicherer. Eines der Angebote von Swiss Re ist die Global Motor Risk Map. Dieser Service erstellt Vorhersagen für Unfallrisiken mit hoher geografischer Granularität basierend auf externen Daten wie Bevölkerungsdichte oder der Struktur des Straßennetzwerks. Dies ermöglicht dem Erstversicherer als Kunde seine Geschäftstätigkeit aufgrund konkreter Vorhersagen zu den Auswirkungen von geplanten Maßnahmen zu steuern. Der Vortrag fokussiert auf „Lessons Learned“ aus der Entwicklung des Produkts in den Bereichen: Aufbereitung der externen Daten, Validierung der Vorhersagen sowie Einbettung in die kundenseitige Prozesslandschaft inklusive automatischer Aufdatierung der Vorhersagen.

Dr. Christian Elsassser, Manager Data Analytics, Swiss Reinsurance Company Ltd

Case Study:

Forecasting in Shell Treasury

Being able to predict customer demand for a particular product and have that product available for sale when required is a fundamental get right for any supply chain. But what if everything we try doesn’t work, and in many cases appears to make the forecast work? Alex gives an account from Shell’s Lubricants Supply Chain on the journey taken by the newly formed Central Forecasting Team in an effort to turn-around and improve a failing metric (Forecast Accuracy) that was being blamed for a wide range of organisational pain. Alex touches on the various ideas that didn’t work (from flat targets, to simply manipulating the data) and the affect these had, before moving on to the later ideas, and career risk, that began to move the needle in the right direction. Key take always include plenty of things to avoid, some useful tools such as segmentation, and the simple question to ask when nothing else works.

Alex Hancock, Head of Treasury Analytics, Shell Oil Company

Case Study:

Predictive Analytics… and Unpredictive Analytics – Strategy in a VUCA World

Political bombshells, unimaginable terrorist attacks, epidemic outbreaks, natural disasters, technological disruptions, fractured markets, transitory advantages, multifarious competitors, increasingly demanding customers and fickle consumers. Thanks to the military, we have a useful descriptor for the conditions and environment these drivers create; ‘VUCA’ – Volatility, Uncertainty, Complexity and Ambiguity… a combination of the magnitude and speed of change, the lack of predictability and prospect of surprise, the multitude of forces and confounding issues, and the lack of ‘one right answer’ or single course of action. Yet against this backdrop many organisations are still carrying the early, often ill-formed, baggage of implicit promises and expectations of Big Data and Analytics, and the mindset of operating in more stable conditions. And that’s before we get to sentiment mining, machine learning, edge analytics and the like. There is no question about the power, pervasiveness, further potential and applicability of predictive analytics. But it is at the edge of this applicability that things get interesting; where a VUCA environment lays cognitive traps for those focused on ‘getting to the right answer’ rather than ‘asking the right questions’. In the context of strategy development, deployment and delivery in the real world, this edge – between prediction and insight, between extrapolation and choice, between algorithm and decision – is critical. This session explores some of the typical problems and hidden traps in creating strategy against a VUCA backdrop, considers the tensions that exist across our unavoidably uneven knowledge of the world, offers some mental models to help leaders grapple with this wicked problem, and to help you drive the most impact and value in support of your strategy.

Chris Turner, Co-Founder, StrataBridge

Case Study:

Predictive Analytics for Vehicle Price Prediction – Delivered Continuously at AutoScout24

AutoScout24 is the largest online car marketplace Europe-wide. With more than 2.4 million listings across Europe, AutoScout24 has access to large amounts of data about historic and current market prices and wants to use this data to empower its users to make informed decisions about selling and buying cars. We created a live price estimation service for used vehicles based on a Random Forest prediction model that is continuously delivered to the end user. Learn how automated verification using live test data sets in our delivery pipeline allows us to release model improvements with confidence at any time.

Christian Deger, Architect, AutoScout24
Arif Wider, Consultant Developer, ThoughtWorks

Case Study:

Nach der Wahl ist vor der Wahl: Predictive Analytics für Wahlprognosen

„Bekommt Rot-Rot-Grün eine Mehrheit im Bundestag?“, „Bleibt Angela Merkel Bundeskanzlerin?“ oder „Wird die AfD die drittstärkste Partei?“. Diese und weitere Fragen tauch(t)en im Vorfeld der Bundestagswahl 2017 auf. Konventionelle Wahlprognosen der Wahlforschungsinstitute liefern jedoch keine Antwort darauf mit welcher Wahrscheinlichkeit solche Ereignisse eintreffen werden, sondern nur eine Stimmanteil-Prognose für die Sonntagsfrage, eine imaginäre Bundestagswahl am nächsten Sonntag.
Wir nutzen die verfügbaren Wahlprognosen und historische Daten um mithilfe von Predictive Analytics konkrete Wahrscheinlichkeitsaussagen über oben aufgeführte Fragen zu treffen. Nate Silver, Buchautor und Statistiker, hat es für die US-Präsidentschaftswahlen in seinem Blog „FiveThirtyEight“ vorgemacht. Seit März 2017 bietet INWT Statistics eine methodisch vergleichbare und wöchentlich aktualisierte Wahlprognose für die Bundestagswahlen an. Neben der Vorstellung der Prognose und der zugrundeliegenden Methodik, wird die Gelegenheit genutzt unsere Prognosen rückblickend zu evaluieren.

Dr. Marcus Groß, Senior Data Analyst

Case Study:

How Predictive Analytics Helped Huawei to Improve Network Performance

Network switch resembles in many ways to traffic signs in a junction, telling the vehicles where to turn to in order to reach a specific destination. Open virtual switch (OVS) is a piece of software that emulates a network switch. It is powerful in that it enables software defined networking (SDN) applications – much more dynamic and flexible compared to traditional networking. However, as a software-based solution, the OVS performance is sensitive to the memory cache optimization in the compute host it is utilizing. In this talk Aviv Gruber will show via an illustrative example of cloud networking how a feature engineering of new scores improves the network performance of the OVS, in particular latency reduction of up to 23% and throughput improvement of up to 33%.

Dr. Aviv Gruber, Data Scientist, Huawei Technologies Co.

Case Study:

Delivery in a Land of no Post Codes

This presentation will give an insight into how we helped one of the largest logistics company in the Middle East to automate the delivery of their shipments using machine learning. The problem was particularly interesting because there are no postcodes in the region. Instead our client would have to make a phone call to each customer and get a description of the address instead, e.g. house with brown door around XYZ roundabout. This process became more difficult because of the cultural dynamics existing amongst the demographics.
This presentation will describe how we leveraged the GPS data the client had to predict the delivery location using clustering and classification algorithms and how we operationalised the model in order to drive actions in real time.

Megha Agarwal, Data Scientist

Predictive Alerting im Kreditmanagement am Beispiel der Concardis GmbH

Die herkömmlichen quantitativen Finanzanalysen unterliegen hinsichtlich Ausfallfrüherkennung mitunter eheblichen Zeitverzögerungen. Aus diesem Grund kommt qualitativen Faktoren und hierbei insbesondere öffentlich verfügbaren Krisensignalwerten eine erhebliche Bedeutung im Credit Risk Management zu. Keyword-Alerting ist hierbei jedoch heute noch viel zu oft statisch. Begriffslisten werden manuell im Web gesucht und liefern unzureichende Ergebnisse. Hohe False-Positive-Ratios beanspruchen erhebliche Ressourcen im Credit Risk Management. Daneben mangelt es der statischen Listenabfrage an Flexibilität hinsichtlich zielgerichteter Erfassung der Kreditrisiken. Gemeinsam mit Concardis hat DATAlovers eine Lösung entwickelt, bei der ausgehend von einer Ereignishistorie und von Schlüsselwörtern die relevanten Konstellationen und umgebenen Inhaltsfragmente über maschinelles Lernen identifiziert werden. Fortan werden diese Muster ebenso als Grundlage für ein Alerting herangezogen. Gleichzeitig lernt der Alerting-Mechanismus aus den Fehlalarmen und schärft die Güte des Alerts. Ziel aus Sicht von Concardis: den manuellen Arbeitsaufwand reduzieren und intelligent über mögliche Veränderungen informiert zu sein.

Jan Marcinkowski, High Risk Merchant Underwriting & Credit Risk Management
Andreas Kulpa, COO, DataLovers AG