All Times EDT
Keywords: human-in-the-loop, time series, scenario forecasting, visualization
Time series forecasting is one of the most prevalent business use cases in data science. Although the academic research focuses more on automated time series forecasting methods, in practice the business decision-making process often requires the interpretability and flexibility to incorporate domain knowledge and scenario analysis for future events to adjust forecasts. To address similar challenges for capacity planning in Cloud Infrastructure Management at Salesforce, we propose the human-in-the-loop framework for timely scenario forecasting which leverages the impact inference of future change points and interactive visualization. This human-in-the-loop approach reduces cost from over capacity provisioning, mitigates risk to our cloud infrastructure by ensuring proper capacity supply, and saves the time spent on manually scenario forecasting by capacity planners.