Bestätigte Sessions
Berlin, 16.- 17. November 2020

Churn Prediction Model and Deployment for Newspaper Subscribers for Mittelbayerischer Verlag

Churn Prediction Model and Deployment for Newspaper Subscribers for Mittelbayerischer Verlag

Summary:

Germany has a printed newspaper circulation of 15 million. We have developed a churn prediction model for Mittelbayerischer Verlag, a large regional newspaper client – first for print/offline products then for online products for 100.000 newspaper subscribers. From business question over data understanding and analysis, to model selection, training, hyperparameter optimization to deployment we have walked through the complete process with the client. Predicting with high quality is hard due to the few features on the actual usage of the print subscribers. Most features are static and they needed to transfer features from one subscriber to peer-groups. The developed solution is based on multiple Gradient Tree Boosting models that each work on different prediction timeframes. They have a lot of lessons-learned due to the large number of features engineered, working with a large group of business people on the client side, going from business question to final deployment in the cloud.

Improving Sales Forecasting for Unilever Through Deep Learning based Object Detection for Ice Cream

Improving Sales Forecasting for Unilever Through Deep Learning based Object Detection for Ice Cream

Summary:

Out-of-stock situations (OOS) are undesirable for brands especially in the ice cream business. Traditional prediction (forecasting) models which depend largely on historical sales data tend to fail while predicting optimal quantity of ice cream stock for various retailers. This is evident as the OOS situation is quite common especially in the summers. One of the reasons is that the models generally don’t use past OOS events to correct themselves. This is simply because gathering OOS data automatically is a challenge. The key to identify OOS in ice cream cabinet is to determine the depth up to which an ice cream basket is filled with the SKU. Mobisy Technologies has devised an automated way to determine the OOS by leveraging our state-of-art deep learning based image recognition technology 35 hawk. The algorithm has been tested in retail stores across Bangalore, India with following outcomes: 80% of OOS events avoided due to better stock predictions. SKU Detection Accuracy: 95%. Depth Detection Accuracy: 90%. This approach is generic and can be extended to dairy, frozen foods which are kept in horizontal refrigerators (cabinets). In this presentation Rohit intends to share their work and results with respect to the Indian market.

Ad Optimization with Multiple Data Sources and Predictive Analytics for Fandom

Ad Optimization with Multiple Data Sources and Predictive Analytics for Fandom

Summary:

How to align understanding of data across whole company? Which data source should we trust (and why) when they send contradictory messages? Finally, how do we reach reliable predictions everyone believes in and wants to use? These are the questions we need to answer even before employing predictive analytics, especially in cross team ads environment. Based on real use cases, Martyna and Andrzej will show how to work with multiple data sources (with the use of our open source Python library – sroka) and how to lead to an agreement between people who believe in different data truths to make predictive analytics happen.

Sometimes Algorithms Matter: When to Bring out the Big Guns

Sprecher:

Dean Abbott

Sprecher:

Dean Abbott

Sometimes Algorithms Matter: When to Bring out the Big Guns

Summary:

For most data science applications, the algorithms do not matter nearly as much as data integration and data preparation. But sometimes, we really need to bring out the big guns: the most sophisticated and complex algorithms to solve difficult problems. This talk will describe situations where complex algorithms matter and when one might want to choose Random Forests, XGBoost or Deep Learning networks to solve those difficult problems.

An Agile Approach to Data Science Product Management

An Agile Approach to Data Science Product Management

Summary:

There are few frameworks on effective AI Project Management. Industry-standard frameworks for data analysis projects, like CRISP-DM, exist but none are effective for managing the development of AI products from deployment to production. The result is that many data science teams are focused on outputting one-off analytical projects, rather than building long-term, maintainable products that directly drive business processes and goals. In this talk Ben proposes an effective project management framework for building production data science systems based on his experience constructing such systems at large tech companies, late-stage start-ups, and early-stage ventures. Specifically, he proposes that a blend of CRISP-DM and the Extreme Programming agile development methodology can serve as a starting point for building organization-specific data science product development processes.

Applying Machine Learning in Order to Increase Revenue for Groupe Casino

Applying Machine Learning in Order to Increase Revenue for Groupe Casino

Summary:

A new retail law saw the day on 2019, one of the rules is the limitation of the number of discount operations done between retailers and food suppliers, which raised the following questions: Which supplier should Groupe Casino do promotional operations with? When would it be done? And under which formula? Answering those questions is essential to maintain or even increase the revenue, so some reliable decision making tools must be established. By turning those question to data investigation and applying machine learning state of the art algorithms, Omar got access to actionable information like a reliable estimation of the revenue of a promotional operation that helped them choose the supplier and the period.

How to Remove Bias and Make Machine Learning Models Fair & Discrimination-Free at the Example of Credit Risk Data

How to Remove Bias and Make Machine Learning Models Fair & Discrimination-Free at the Example of Credit Risk Data

Summary:

There have been multiple instances when a machine learning model was found to discriminate against a particular section of society, be it rejecting female candidates during hiring, systemically disapproving loans to working women, or having a high rejection rate for darker color candidates. Recently, it was found that facial recognition algorithms that are available as open-source have lower accuracy on female faces with darker skin color than vice versa. In another instance, research by CMU showed how a Google ad showed an ad for high-income jobs to men more often than women. Using credit risk data where Publicis Sapient wanted to predict the probability of someone defaulting on a loan, they were able to shortlist features that were discriminatory in nature.

How Did Predictive Analytics & Attribution Help Fingerspitzengefühl To Optimize Cross-Channel Marketing Campaigns

How Did Predictive Analytics & Attribution Help Fingerspitzengefühl To Optimize Cross-Channel Marketing Campaigns

Summary:

Fingerspitzengefühl (FSG), an agency, supports international companies in successfully establishing and expanding their online business in Germany. With FSG, Haensel AMS has developed an analytics & attribution system that connects all data sources at individual user touch point level. With the help of a novel attribution model, the performance of various marketing activities/campaigns and their interactions can be determined and thus the profitability of the marketing budget in the multi-channel mix can be optimized. In the talk Alwin and Radboud will introduce and explain the data basis, the concepts of the attribution & analytics models, the design of the dashboard, as well as some FSG use cases.

Stock Price Prediction and Portfolio Optimization Using Recurrent Neural Networks and Autoencoders

Stock Price Prediction and Portfolio Optimization Using Recurrent Neural Networks and Autoencoders

Summary:

Financial time series forecasting is a challenging problem. Deep learning approaches, such as recurrent neural networks (RNNs), have proven powerful in modelling the volatility of financial stocks and other assets, as they are able to capture non-linearities in sequential data. Recent studies have shown that RNNs have surpassed well-known autoregressive forecasting models (Siami-Namini, 2018). Besides forecasting the next value of a stock, mVISE is also interested in creating an optimal portfolio. Deep portfolio theory (Heaton et.al.,2018) uses autoencoders to model the non-linearity of the time series to accurately predict returns. Julian will extend this approach by first performing a 5-day ahead forecast and then train an autoencoder model to construct an ideal portfolio that incorporates previous and future stock market information. Attendees will be provided with a theoretical understanding of how RNNs and autoencoders work and how to apply them on multivariate timeseries forecasting and portfolio optimization problems. He will backtest the new model on financial stock data. Finally, the session will end with presenting an overview of key challenges and current research topics within that field.

How to Classify News Articles at Scale for Upday

How to Classify News Articles at Scale for Upday

Summary:

Upday serves over 85K news articles to millions of users across Europe every day. This means they process a lot of textual data in many languages and contexts. In order to connect people with the right content upday needs to know what the articles are about – they need to classify them. The goal of their latest project was to replace a rule-based classification system with a machine learning model. The new model should be fast, easy to scale to further markets and perform well with little training data. In this talk, Helena and Malgorzata will present their journey to finding an algorithm that meets these expectations.

Newsletter

Fundiertes Wissen ist die Basis für alles!

Melden Sie sich für den Newsletter an und erhalten Sie:

  • 10% Rabatt auf Ihr Ticket
  • Einblicke, Interviews, Tipps, Neuigkeiten und vieles mehr
  • Erinnerungen an Preissenkungen