The bias in reversing the Box–Cox transformation in time series forecasting: An empirical study based on neural networks

Costa, Alexandre Fructuoso da; Crepaldi, Antonio Fernando

doi:10.1016/j.neucom.2014.01.004

Cited by 5 publications

(2 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Non-stationary data in variants can be performed Box-Cox transformations [14], [18]. The Box-Cox transformation equations are as follows [19], [20], [21].…”

Section: Stationaritymentioning

confidence: 99%

Arima Model of Outlier Detection for Forecasting Consumer Price Index (Cpi)

Imron

Utami

Khaulasari

et al. 2022

BAREKENG: J. Math. & App.

View full text Add to dashboard Cite

The Consumer Price Index (CPI) is a indicator used by Badan Pusat Statistik (BPS) which describes the average change in the prices paid by urban consumers for a market basket of consumer goods and services in a certain period. The case on Consumer Price Index (CPI) of Probolinggo City, if the Consumer Price Index (CPI) increase then describe inflation occurs and conversely. The Consumer Price Index (CPI) of Probolinggo City increase is not fixed. This study is to forecast the Consumer Price Index (CPI) that the results can be used as one of the considerations in carrying out economic development in the future. Research focused on the data of Consumer Price Index (CPI) of Probolinggo City from January 2014 to April 2022. Methodology implemented in this study is Autoregressive Integrated Moving Average (ARIMA). Result show that ARIMA without an outlier was the best model for predicting Consumer Price Index (CPI) of Probolinggo City for the next 8 months. This model shows the value of MAPE is . The value of forecasting results in each month has decreased and increased not so significantly where in May 2022 the forecasting value was 108,391 then in June 2022 the forecasting value became 108,411 and so on until December 2022 the forecasting results using ARIMA model of 107,845.

show abstract

“…Non-stationary data in variants can be performed Box-Cox transformations [14], [18]. The Box-Cox transformation equations are as follows [19], [20], [21].…”

Section: Stationaritymentioning

confidence: 99%

Arima Model of Outlier Detection for Forecasting Consumer Price Index (Cpi)

Imron

Utami

Khaulasari

et al. 2022

BAREKENG: J. Math. & App.

View full text Add to dashboard Cite

show abstract

“…The BCT is often applied to time series data to obtain variance stability, Gaussianity of the distribution function, and additive seasonality (Fructuoso da Costa and Fernando Crepaldi, 2014).…”

Section: Figure 2 Turkey's Hazelnut Export Datummentioning

confidence: 99%

Forecasting Turkey's Hazelnut Export Quantities with Facebook's Prophet Algorithm and Box-Cox Transformation

Aytaç

2021

ADCAIJ

View full text Add to dashboard Cite

Time series forecasting methods are used by an evolving field of data analytics for the prediction of market trends, sales, and demands. Turkey is the major producer of hazelnut in the world. If Turkey wants to continue its domination of hazelnut and protect the price-setting role, time series forecasting methods could be key factors accordingly. There are a few studies that focused on time series forecasting of hazelnut export quantities of Turkey, and this study uses a recently developed algorithm and implements a power transformation to increase the forecast accuracy. The presented research aims to forecast Turkey’s hazelnut export quantities for the coming 36-months starting from June 2020. The forecasting process was conducted with the help of Facebook’s Prophet algorithm. To improve the forecast accuracy, a Box-Cox power transformation was also implemented to process. To find out the stationarity and periodicity of the data set, the Augmented Dickey-Fuller test and autocorrelation function was applied to the time-series data. The Prophet algorithm, with Box-Cox transformation, projected the hazelnut export quantity could be over five hundred thousand tons from 07/2020 to 06/2023. The export quantities were in an increment trend, and the slope of the trend has increased since June 2008 by 0.66 % per month. The Prophet algorithm also revealed the seasonality of the data set, and the export amounts indicate monthly oscillations. The monthly export volumes start to increase and reach their peak value in October because August is the time for the harvest of hazelnuts in Turkey.

show abstract

A single-tree additive biomass model of Quercus variabilis Blume forests in North China

Zheng

Mason

Jia

et al. 2015

Trees

View full text Add to dashboard Cite

Key message Fitting and comparing three sets of compatible biomass models for prediction of biomass or carbon stocks of natural and planted Quercus variabilis Blume forests.Abstract To make the sum of estimated values from biomass models of various components of a tree equal to estimated tree total biomass for Quercus variabilis Blume (cork oak) forests in North China, single-tree compatible biomass models were developed. 100 trees from 100 plots in North China were felled to obtain biomass of aboveground components, and roots of 19 of those trees were extracted for measurement of root biomass. After Box-Cox transformations of variables, two sets of independent component biomass models with a dummy variable in models to define stand origin were separately built using linear mixed effects analyses (one set of models with site as a random factor; the other set without any random factor). Then three methods were compared to force compatibility of the biomass models: sums of linear mixed effects models, sums of linear models, and simultaneous equation fits based on linear models. Model parameters were estimated by ordinary least squares (OLS) or seeming unrelated regression procedures (SUR). Coefficients of determination (r 2 ), root mean square error (Rmse), residuals plots and histograms of residuals indicated that models fitted with sums of linear mixed effects models were the least biased and most precise at estimating total above-ground biomass.Further testing for the linear mixed effects models with jackknife validation and prediction sum of squares (PRESS) statistics indicated that the compatible biomass models can be used to predict biomass or carbon stocks of cork oak forests in North China within specific tree diameter at breast height and height ranges.

show abstract

The bias in reversing the Box–Cox transformation in time series forecasting: An empirical study based on neural networks

Cited by 5 publications

References 28 publications

Arima Model of Outlier Detection for Forecasting Consumer Price Index (Cpi)

Arima Model of Outlier Detection for Forecasting Consumer Price Index (Cpi)

Forecasting Turkey's Hazelnut Export Quantities with Facebook's Prophet Algorithm and Box-Cox Transformation

A single-tree additive biomass model of Quercus variabilis Blume forests in North China

Contact Info

Product

Resources

About