Skip to content

fialhocoelho/ARIMA

Repository files navigation

ARIMA

Generic ARIMA analysis for time series implemented from https://towardsdatascience.com/time-series-in-python-exponential-smoothing-and-arima-processes-2c67f2a52788

Related works:

  • Kumar, Prashant. "Forecasting Cloud Resource Utilization Using Time Series Methods." (2018). http://www.diva-portal.se/smash/get/diva2:1273037/FULLTEXT01.pdf (ARIMA + ES) > ARIMA FFNN > (ARIMA + ES)

    • Good description of data preparation. We can adapt our data to use the same configuration without too many problems Just uses the CPU measurements
  • Paul Newbold and C W. J. Granger. “Experience with Forecasting Univariate Time Series and the Combination of Forecasts”

    • This paper suggested that a combination of individual forecasting models performs better than any individual forecasting.
  • Islam, S., Keung, J., Lee, K., & Liu, A. (2012). Empirical prediction models for adaptive resource provisioning in the cloud. Future Generation Computer Systems, 28(1), 155-162.

    • “... Individual samples may not be a representative of the true resource utilization level… ”

Data prep:

  • “This kind of data tends to have a trend” - Kumar, Prashant. "Forecasting Cloud Resource Utilization Using Time Series Methods." (2018).
  • The same data gaps for CPU and Memory
  • The CPU and Memory are High correlated
  • It's a bad idea to use CPU and Memory together as a hyperparameters, but it’s a good idea to test them individually to check the true representativeness

Completed tasks:

  • The data gaps are filled using interpolation techniques
  • A standard for all the files considering a hourly frequence

Time series Analysis:

  • A trend (upward or downwards movement of the curve on the long term)
  • A seasonal component
  • Residuals

About ARIMA:

  • ARIMA models should be used on stationary data only.
  • To obtain the best performance of model, we need to make the data stationary
    • De-trending
    • Seasonal adjustment
    • Transformation
    • Smoothing
  • Time series with trends, or with seasonality, are not stationary

Smoothing methods:

  • Smoothing methods work as weighted averages. Forecasts are weighted averages of past observations.
  • Simple Exponential Smoothing
    • Few data points, Irregular data, No seasonality or trend.
  • Holt’s Linear Smoothing
    • Trend in data, No seasonality.
  • Holt’s Damped Trend
    • Data has a trend
  • Exponential smoothing (ES)
    • One of most flexible methods (related to time series patterns)