TY - JOUR
AU - Booranawong, Apidet
AU - Booranawong, Thitima
AU - Wattananavin, Thradon
AU - Nikhom, Ruamporn
AU - Auysakul, Jutamanee
PY - 2021/06/18
Y2 - 2024/03/04
TI - Analysis of AHW and EAHW Time-Series Forecasting Methods: A Mathematical and Computational Perspective
JF - Naresuan University Engineering Journal
JA - NUEJ
VL - 16
IS - 1
SE - Research Paper
DO - 10.14456/nuej.2021.2
UR - https://ph01.tci-thaijo.org/index.php/nuej/article/view/241700
SP - 7-13
AB - <p class="abstract"><span lang="X-NONE" style="color: black; font-weight: normal;">Recently, an Extended Additive Holts-Winter method (EAHW) as a modified version of the well-known Additive Holt-Winters method (AHW) has been introduced. Many works in the research literature applied such a method to forecast time-series data, and results indicated that the EAHW provided optimal forecasting accuracy. However, we have found that there is no work in the literature analyzing the AHW and the EAHW in terms of the mathematical operations, which refer to the computation time and complexity of the methods. Since the AHW and EAHW can be implemented on hardware platform for using in many related systems and applications, in this paper, analysis of both methods from a mathematical and computational perspective is presented, where the mathematical operations of the EAHW compared with the AHW is provided. We show that the major difference between the EAHW and the AHW is the calculation of the level of time-series data, while the calculation of the trend and the seasonal factors are the same. With such a difference, the EAHW provided different possible solutions for forecasting. Also, in the worst case scenario, the EAHW has more multiplication operations (i.e., operation ×) than the case of the AHW method (but it is not too high); 2<em>N</em> for the AHW and 3<em>N</em> for the EAHW, where <em>N</em> is the number of data samples. By our finding, the EAHW is one of the appropriate forecasting methods for implementation (i.e., both on computers and embedded hardware platforms), since it provided good accuracy and computational complexity.</span></p>
ER -