Abstract: Time series modeling is a challenging task with a unique set of problems. Time-series data often comes in very large quantities of many different data streams which constantly change and evolve over time. While many modeling techniques already exist, they often have major limitations when it comes to scalability, agility, explainability and the effort involved to achieve good accuracy.
It’s been 50 years since ARIMA and Holt-winters models were introduced to the public and ever since data scientists from all over the world have tried to make this technique, other variants and even completely different modeling techniques (Neural Nets, LSTM, XGBoost,..) work on their time-series data but very often it doesn’t and definitely not at scale.
Time series require a different approach, with efficient and effective feature engineering at its core. That’s what the game-changing Tangent Information Modeler (TIM) is about. Using a breakthrough architecture based on Information Geometry this time series ML co-pilot automatically generates new feature transformations for you, making it a multivariate modeling technique that can handle a wide range of time series use cases with award-winning results and incredible performance.
During this workshop, you will build complex time series forecasting and anomaly detection models from the ground up, perform feature engineering and selection, assess the accuracy, and utilize the ModelZoo browser and root cause analysis functionalities to investigate the outcomes.
Bio: Philip Wauters is Customer Success Manager and Value engineer at Tangent Works working on practical applications of time series machine learning at customers from various industries such as Siemens, BASF, Borealis and Volkswagen. With a commercial background and experience with data engineering, analysis and data science his goal is to find and extract the business value in the enormous amounts of time-series data that exists at companies today.