Abstract: Regardless of their concrete application the primary goal of forecasting systems is to produce the most accurate forecast possible. However, while beating benchmarks is important, a forecast useable in business processes additionally needs to fulfill many more criteria, which significantly increases the complexity of real-world solutions. The business requirements can vary depending on both the type of forecast and its goal.
They can reflect strategic business goals (e.g., inclusion of predefined target volumes to which the forecast has to adhere to some extent), ensure the consistency of the forecast model (e.g., enforcing limited volatility between forecast runs), refer to process challenges (e.g., the need to limit pipeline complexity to produce results quickly), or aim at facilitating trust with the end user (e.g., requiring every prediction to be explainable).
In this tutorial we share best-practice methods for building highly accurate open source ML frameworks that have emerged from numerous industrialized forecasting systems in a multinational and interdisciplinary company:
1) The best-in-class ML modeling approaches for high out-of-the-box accuracy in different data scenarios;
2) Automated feature generation methods for improving accuracy further;
3) The least intrusive forecast adjustment methods that facilitate use of the product in business; and
4) Implementations of explainable AI that can enable end user trust in ML methods.
After attending this session participants will have extended their knowledge on
1) how to formulate time-series forecasting as a Machine Learning problem;
2) which modeling approaches work well for which kind of data and time constraints;
3) which feature engineering approaches are really increasing accuracy (including open source Python libraries for automated feature generation);
4) potential business requirements above accuracy and how to tackle them with techniques such as asymmetric loss functions or postprocessing of predictions;
5) using available explainable AI technology to convince end users of the sanity of models; and
6) what the actual business benefits of automated forecasting solutions are.
Bio: David Koll is a Senior Data Scientist at Continental Tires, Germany. He holds a PhD in Computer Science from the University of Göttingen with research visits to the University of Oregon (USA), Uppsala University (Sweden), and Fudan University (China). Most of his academic work was involving analyses of social media. Since joining Continental in 2018 he has developed different analytical solutions that are now running in production, with a focus on both forecasting and Industry 4.0.