Fig. 4
From: Efficient resource provisioning for elastic Cloud services based on machine learning techniques
Autocorrelation analysis of input training data. To choose an appropriate lag period, we analyzed the seasonal patterns of the time series, by measuring the autocorrelation of the input training data for different lag values, as shown in this figure. This was achieved using the Excel autocorrelaction function (rk, where k is the lag interval), which is computed, as rk=sk/s0, where sk is the autocovariance function at lag k, and s0 is the variance of the time series. As we can see in this figure, the input data exhibit a clear autocorrelation for a lag interval of 24 h. For this reason, the chosen lag period was T=24 hours