Applying Machine Learning Models to Predict Operational Demand Using Cloud-Native Data Platforms

Abstract

Operational demand forecasting is critical for effective resource allocation in any organization. However, keeping accurate forecasts can be difficult due to factors such as seasonality, trend, and stochasticity. Although multiple methods exist to tackle this problem, considering Machine Learning (ML) approaches is essential to allow for rapid inclusive assimilation and integration of historical data. By harnessing the capabilities of a cloud-native data platform and optimally addressing the various aspects of a typical ML workflow (from data acquisition to experiment deployment), it is important to provide evidence of the accuracy of these methods for forecasting across horizons of small (operational) to intermediate length. Indeed, even if a method is poor at predicting the future, it may still be good enough for deployment in the present.