ABSTRACT
Time series analysis is widely used to forecast the future events. It can learn and extract the patterns from the data indexed in time. The time series models require periodic parameter learning to capture the behavioral change of workloads over time. The dynamic systems such as cloud servers usually see the frequent changes in the data pattern. Thus, time series analysis models may not forecast the cloud workloads with reasonable accuracy. In this chapter, an error prevention scheme is discussed which learns the error trend in the recent forecasts and this value is further used to improve the next forecasts. Thus, it addresses the requirement of periodic learning in a time series model. The error prevention scheme is integrated with the time series models and their performance is compared with their counterparts. In this chapter, the auto regressive moving average, autoregressive integrated moving average, and exponential smoothing models are used for the experimentation purpose. The error preventive versions of these models are discussed and their workload forecasting abilities are compared and analyzed critically.
