• Design an ML production system end-to-end: project scoping, data needs, modeling strategies, and deployment requirements
  
• Establish a model baseline, address concept drift, and prototype how to develop, deploy, and continuously improve a productionized ML application
  
• Build data pipelines by gathering, cleaning, and validating datasets. Establish data lifecycle by using data lineage and provenance metadata tools
  
• Apply best practices and progressive delivery techniques to maintain and monitor a continuously operating production system
  

Description

• Conceive and implement time series forecasting to predict wind patterns in the energy sector
• Consult on time series analysis methodology

Practical work

• Conceive and implement following forecasting models (ARIMA, SARIMA, grid search ARIMA, supervised learning for time series forecasting)
• Consult the client on the analysis and hands on implementation of methods such as stationary and non-stationary time series, detrending time series, seasonality in time series, Augmented Dickey-Fuller test
• Consult on time series analysis methodology for uncertainty assessment
•  Consulting in Resampling methods such as time series bootstrapping
• Simulation of time series within a given range of requirements 
• Train software developers and consult them in time series methodology 

Description

• Conception and Implementation Machine Learning Models to predict if an applicant will convert to student
• Advise and provide new ideas for the business units in the subject areas of Advanced Analytics, Machine Learning and AI
• Collaborate closely with the business units to create well-fitting Data Science and AI Solutions

Practical work

• Suggest and implement new features to improve the performance of the Data Science Model
• Implement deep learning models
• Introduce the Statistical modeling of time series to improve the quality of the prediction

• Development of interactive Dashboards to assist the Risk Management Departments in the visualization of Key Risk Indicators
• Digitalization of the existing manuall processes and improving the ability of the Business Experts to analyse large amounts of data
• Conceiving and implementing the necessary data architecture for the improvement of the performance of the dashboard with very large amounts of data
• Assist in the screening of optimal technology stack and suggestions for an optimal choice for the project
• Implementing and outlier robust alternative for time series forecast

• Conception and Implementation dashboards with R Shiny to analyse large amounts of data
• Suggest and implement an alternative data layer to improve the performance of the required dashboards
• Implementing data loading procedures and performing data migrations
• Statistical modeling of time series to improve the quality of the prediction

Practical Work 

• Conception and Implementation of the methods with R
• Expert Knowledge Transfer in the approaches to automatic time series analysis
• Statistical modeling and application of the methods on real life data

Use Case ?Backwelt?

Practical Work

• Establish concepts and organization structure for data science
• Establish requirement management guidelines for data analysis
• Using Advanced Analytics insights to optimize processes
• Solving complex business relevant problems using Predictive Analytics
• Use Case: Apply Machine Learning to predict whether or not a customer will call the hotline within 2 weeks of his previous call to optimize customer satisfaction. Use Deep Learning on huge amount of technical data on the hardware of the client and more additional data from customer behavior.

Parctical Work

• Involved in the RfP (Request for Provider) Predictive
• Use case ?Weigh and measure?
  • Conception and implementation of a classification method to determine the subset of all shipments to be weighed and measured with the goal to weigh at most 10% of the shipment and still detect 50% of all deviations in weight or volume
• Use case ?Predict booking?
  • Apply time series analysis to forecast the overall weight and volume on a plane several day before departure by using historical data on booked shipments
• Use case ?Quotation Analysis?
  • Statistical analysis of special discount offers on specific flights (the so called quotations) to determine the probability that the offer was not accepted by the client due to the proposed price

Practical Work

• Perform the analysis with SAS and R integration in SAS
• Cross reference the performances of predictive Analytics methods such as market basket analysis with those from the SAP PAL (Predictive Analytics Library) of SAP HANA

Practical Work 

• Earlier detection of outliers and level Shifts
• More accurate predictions of the time series using the knowledge in the detected structural breakpoint in real time
• Programming of statistical packages in R (Ongoing)
• Scientific Publication of the findings (Ongoing)

Practical Work 

• Build a stable systems which uses a vector autoregressive moving average (VARMA) model to analyse and predict multivariate time series after analysing co-integration
• Collaboration with computer scientists to build a graphic user interface
• Test and monitor the goodness of fit the statistical model

Practical Work

• Improve the method of Harrison and Stevens (1971) for time series online monitoring
• Compute a posteriori probabilities of occurrence of a state change at each point in time
• Find a predictive analytics method that uses the a posteriori probabilities and further information to optimally classify the state changes at each point in time Results
• A linear discriminant analysis (LDA) rule to classify the state changes in real time
• A second rule that updates the classification one point after the first classification using additional information available
• Reduction of classification error rate by more than 50% w.r.t standard methods for classification
• Improvement of the accuracy of predictions of the time series

Practical Work

• Develop a new robust Shapiro-Wilk test of normality in the presence of outliers
• Mathematical derivation of the asymptotical null distribution of the new test for the use of critical values
• Derive a robust test that outperforms the most robust test of normality in the presence of outliers

Practical Work

• Collaborate with computer scientist to build a user graphic interface
• Conceive and implement a vector autoregressive moving average model to capture the cross correlation between interrelated time series with cointegration
• Compute predictions for the whole system of time series (< 6 Time Series)

Practical Work

• Collaborate with computer scientist to build a user graphic interface
• Cross correlation and causal analysis for two time series
• ARMA and ARMAX modeling to capture the correlation between the time series
• Conceive and implement a graphic user interface with incorporated monitor for goodness of fit

Practical Work

• Causal and cross correlation analysis first for milk consumption and butter stock, secondly for milk price and whey powder price
• ARMAX models for milk consumption with butter stock as proxy-variable and for milk price with whey powder price as proxy-variable
• One-step and two-step predictions for both time series

Result

Better prediction results w.r.t the univariate analysis

Practical Work

• Collaborate with computer scientists to build a graphic user interface
• Conceive and implement a system to automatically extract seasonal and trend effects in a time series
• Conceive and implement a system to monitor, control and detect state changes in time series using a state space model with Bayesian parameters