Skip to main content

The highlights of the lecture program on day 2

 

The Awakening of AI

Also on the second day, there were exciting presentations. In his keynote titled "The awakening of AI! Success stories in time series prediction", Dr. Sven Crone went back in history and presented, in a very vivid and entertaining way, the audience with early experiments in machine learning (no, not his own)  in some detail, inter alia, by way of a video from the year 1952. The presentation then focused on more recent projects in which time series data were used to predict sales figures. His examples of application extended from several thousand products in the retail sector in the United Kingdom to consumption of a Belgian brewery's beer.
 

Die Highlights am zweiten Tag der Predictive Analytics World

In his lecture titled "The value of external data for predictive analytics in the B2B environment using the example of Swiss Re" Dr. Christian Elsässer from Swiss Re presented some lessons learned in the development of the global motor risk map. Incorporated here, for example, are data on population, night-light emission (as a proxy for economic activity), weather and information on road networks. Swiss Re offers this map to direct insurers for better assessment of accident risks at the regional level.

In their report titled "Predictive alerting in credit management using the example of Concardis GmbH", Joachim Gaschler from Concardis and Andreas Kulpa from DataLovers presented an alerting solution which helps Concardis as provider of cashless payment services to better assess credit default risks. Involved here are possible reclamations by end customers which can (usually) take place 120 days after a transaction, for example, if a product is not delivered. The alerting solution is created on the basis of Google Alerts using various algorithms which are routinely tested against each other in some cases (word2vec, neural networks and random forest). Concardis accordingly has a kind of early warning system intended for closer consideration by their customers (numbering more than one hundred thousand).

In his presentation titled "Predictive analytics for vehicle price prediction - delivered continuously at AutoScout24", Christian Deger from AutoScout24 described the environment and methodology serving as a basis for the prices which AutoScout24 estimates for its customers. Fascinating here was the technology used to integrate the data science team's analytical model (R-based estimation, random forest) into a continuously updated price estimate for a web application accessible by users of web pages.

Dr. Markus Groß from INWT Statistics presented, under the title "After the election as seen before the election:" Predictive analytics for forecasting election results", an analysis of the German parliamentary elections 2017 based on survey results crawled from survey institutes and Monte Carlo simulations. Involved here was a macro forecast for determining probabilities of certain events regarding the election outcome, rather than the exact results of the election.
 

Deep Dives

In addition to case-study presentations, there were some deep-dive lectures which, however, in some cases had limits imposed on participation or were overcrowded due to the small-sized rooms. Because Predictive Analytics World was held under the theme of data driven business - customer centricity through data jointly with the conferences eMetrics and digital growth unleashed, there were also other presentations from which one could choose.
 

Die Keynote - SpiegelMining

In the final keynote at Predictive Analytics World titled "SpiegelMining - data science at Spiegel Online", David Kriesel presented the results of his analysis of almost 100,000 Spiegel online articles since mid-2014. In addition to a few "fun facts" and descriptive statistics on numbers of articles in the course of time, there were also many colorful insights into the proximity of different topics to each other. Portrayed as tricky was the feature as to whether an article may be commented, and related data over the course of time were displayed. The relative rise in the number of articles allowing comments after the first public lecture by David Kriesel on SpiegelMining in December 2016 was brought into relation here. Though some would say that a contrasting fact is still needed to reveal evidence of the presentation's effect here, perhaps the Spiegel online editorial also had some items of "small data" (in the form of discussions) to contribute to this conclusion.

Overall a very exciting conference!

Dr. Michael Allgöwer
Dr. Michael Allgöwer
Management Consultant
Machine learning has been Michael's field of expertise for some time. He is convinced that high-quality machine learning requires an in-depth knowledge of the subject area and enjoys updating this knowledge on a regular basis. His most recent topic of interest is reinforcement learning.
#MachineLearning #ReinforcementLearning #HumanIntelligence