Predictive Analytics


Luis Guimaraes


Faculdade Engenharia Universidade Porto, WindRel: Estimating Wind Turbine Generators Reliability

EDP Renewables (EDPR) is a leading renewable energy company operating 8.6 GW around the globe (US third largest wind power producer). Its main focus is onshore wind power generation with over 190 wind farms and 6,000 WTG across 12 countries. Wind power has received much attention due to the concerns with greenhouse gas emissions and the world’s installed capacity will double by 2018. However, the euphoria of new installations has faded and operation and maintenance (O&M) costs are becoming a major concern. This project tackles the development of WindRel, a system which dynamically estimates the different WTG components reliability throughout its useful life, considering the effect of internal and external factors. The system is used to estimate O&M costs in current and investment projects given its ability to predict failure rates. WindRel is a step on EDPR’s endeavor of creating an advanced data-based and conditional-monitoring maintenance system to optimize O&M costs.


Luis Guimarães is an Invited Assistant Professor at the Industrial Engineering and Management Department – Faculty of Engineering of University of Porto. He is also a Principal Researcher and member of the coordination board of the Industrial Engineering and Management Research Unit of the Associate Laboratory INESC-TEC. Luis Guimarães’s main area of activity is Operations Research/Computer Science. Most of his research is problem-driven and aims to develop advanced analytic solutions to be applied in real-world problems. In this context, he has collaborated in more than 15 industry-based research and consulting projects with various companies, including Unicer, Europac Viana, BaVidros, Soane and EDPR in the areas of process industry, transportation, retail and energy. Author of several publications in international journals in the field of Operations Research. He has given over 30 oral presentations in international conferences and seminars.

Back to schedule


Jerry Shan


Automatic Identification of Dynamic Drivers for Modeling and Prediction

How can data scientists help decision makers gain better visibility in the complex business world? For example, what key drivers or factors are impacting HP’s revenue in the PC market? What controllable factors can we tune to help the business move in the direction that we desire? Is this presentation, Jerry will show how he explored those questions through one concrete business analytics solution that he and his team created for PPS Finance of Hewlett-Packard Company: Automatic driver identification for revenue modeling & prediction. The analytics solution that Jerry created analyzes the various enterprise data, answers tough questions with rigorous data analysis and models, and provides decision makers with crucial decision support. Based on the analysis and modeling technologies, the solution demonstrates the significant values of business analytics with big data.


Jerry Shan has extensive research and development experiences in data analytics, with a track record of solution creation and technology transfer for HP internal business and external client companies. In addition, Jerry has numerous publications and 24 granted US patents. Jerry has earned his PhD in Statistics from Stanford University in 1995.

Back to schedule

Dirk Van den Poel

Ghent University

Predictive and Prescriptive Analytics in Door-to-Door Sales Companies

In a world characterized by data deluge, taking the best decisions based on the available information is not easy for small businesses. While intuition and business knowledge are often the key driver in these environments, data-driven decision making is increasingly becoming more important in approaching most operational decisions. This is particularly true for sales teams. In order to maximize the revenue they collect, they need to have knowledge on the value of each customer and need to decide whom they will select for each visit and in which sequence they will visit them. The complexity of this task is more difficult than many of them are ready to admit.

For the valuation of customers, the recency effect is often quite strong. However, other variables such as frequency, total monetary value, seasonality and situational variables such as the time of the day can all have a huge influence on the total value of a customer. This makes their forecast on the value unreliable. In the generation of sales routes,  representatives usually rely on their own knowledge of the area, which can be insufficient and outdated. Furthermore, they often focus too much on visiting all customers that are on their way, rather than focusing on the ones that generate profits. This is as they do not take into account the actual cost of visiting them. This results in sales routes that visit too many unprofitable customers in an inefficient way.

Linking the outcomes of statistics, data mining and machine learning algorithms with decision making capabilities from the operations research field, we have developed a two-stage approach that effectively tackles this situation. In a first phase we calculate the potential value of each customer in the database by using analytical customer valuation techniques from the Customer Relationship Management area. We use transactional data, situational data and customer profiles to make our predictions. In the second phase we schedule these clients using state-of-the-art routing algorithms and visualize these routes with a custom developed GIS tool. We compare two different approaches for solving this routing problem and propose an interesting trade-off between computing time and increased profit.

Our framework was tested in collaboration with a door-to-door sales company and gained four significant results: we improved the hit rate of the visited customers, increased the collected revenue per customer, lowered distribution cost and fuel consumption and increased overall profits. While there was initial resistance to this data-driven approach, the results quickly motivated all sales representatives to join in. As a side effect, the retention of their representatives also went up.


Dirk Van den Poel (PhD) is Full Professor of Marketing Analytics/Big Data at Ghent University (Belgium). Since 1999, he teaches courses such as: Statistical Computing, Analytical Customer Relationship Management, Marketing Information Systems/Database Marketing, Big Data, Marketing Models and Marketing Engineering, Marketing 101. He co-founded the advanced Master of Science in Marketing Analysis, the first master in predictive analytics in the world.

His major research interests are in the field of analytical CRM (Customer Relationship Management). His methodological interests include ensemble classification methods, artificial neural networks, reinforcement learning. He has been the supervisor of more than 10+ completed PhDs (10+ more in progress).

He has co-authored more than 70 international peer-reviewed (ISI-indexed) publications. His ISI h-index currently is 20 (i.e., he co-authored 20 Web-of-Science indexed publications with at least 20 academic citations). His Google h-index is 32, with a total of 4,030 academic citations. He has given 100+ talks at national and international business as well as academic conferences, and has chaired several events.

Back to schedule

Yale Zhang

n-Side SA

How Advanced Analytics Can Convert Your Energy Flexibility into Tangible Cost Savings?

With the increasing price and price-volatility of electricity, industrial consumers are facing significant challenges to reduce electricity bill and attain competitive advantage. Fortunately, many industrial sites have unexploited energy flexibilities that can be leveraged to achieve above goals.  Based on our experiences in developing an algorithm to set European electricity price and volume for day-ahead market, a new integrated decision framework, “ENERTOP”, is being developed to help energy-intensive companies make a better decision between production scheduling, cogeneration planning and positioning on electricity market. During this presentation, we will show our approach combining predictive analysis to forecast electricity price with prescriptive analytics to optimize energy supply and demand. Benefits gained from such an analytical approach are demonstrated by a real-world case study in process industry.


Yale Zhang, Ph.D. in control engineering from Tsinghua University, Beijing. He is Consulting Manager at N-Side S.A. His professional interests focus on data-driven modeling, discrete event simulation, mathematical optimization and its applications in process industries such as steel, mining and metal sectors. He is member of INFORMS and International Federation of Automatic Control (IFAC) technical committee on Automation in Mining, Mineral and Metal Processing since 2007.

Back to schedule