Time Series Forecasting and Dynamic Pricing for Cloud Usage

Abstract

This capstone explores different classical time series forecasting models to forecast cloud usage for an Infrastructure as a Service (IaaS) provider. The objective is to provide forecast information to help with capacity planning and propose a pricing model to optimize the capacity and manage revenue. The Mean Absolute Percentage Error (MAPE) performance criteria was compared for all candidate forecasting models to select the most suitable one. Analysis of the data showed a high linear trend in most of the zones, as well as a weekly seasonality. An elastic pricing model was proposed to influence customer demand behaviors to smoothen out capacity during the week. The conclusion is that the demand can be forecasted using a linear model with weekly seasonality. The determination of the most suitable forecasting model and prescribed elastic pricing model will help the sponsor company plan and manage capacity and revenue more optimally.