IEEE Consumer Electronics Magazine - November/December 2021 - 49

most influencing factors. The emission of GHG
results in acid rain and earth's temperature rise,
i.e., global warming. The destabilizing of the
global climate destroys natural ecosystem and
increases natural disasters, such as violent
storms, floods, droughts, etc. The acid rain and
these natural disasters affect the agriculture productivity.
Using the same technique, we
obtained subset for cotton consumption that
includes six influencing factors, wheat production
(wpt), wheat consumption (wct), cotton consumption
(cct), loans (lt), rural population (rut),
and CO2 gas emission (cgt). Meanwhile, the
selected subset for rice consumption affecting
factors are comprised of wheat consumption
(wct), rice consumption (rct), population (pt),
and methane gas emission (mtt).
The relationship concerning different socioeconomic
and environmental factors is evaluated
with respect to the consumption of selected
crops. The Coefficient of Determination R2 is calculated
to check how closely data points are
located to regression line. Factors with highest
R2
are more appropriate because they exhibit
strong correlation. Then these factors are
passed into linear regression and recursive neural
network algorithms to generate various models.
To verify the validity of models, data
splitting of 70/30 is used for training and testing.
Experiments and analysis are carried out in R
Studio, which is an open source IDE, powered by
the R programming language, typically used for
scientific and statistical computations.
RESULTS AND DISCUSSION
In view of the purposes of this work and the
nature of statistical evidence, two methods are
considered for forecasting during data analysis.
Multiple learning models are constructed using
linear regression parallel with recurrent neural
networks to study the relationship between consumption
and its influencing factors, and then
forecast the future consumption. By using the
socio-economic and environmental factors notations
from Table 3, we construct our first model
by introducing the regressing persuasive factors
with AR(1). Here, if all the influences become
input variables, we get the modified linear
regression model together with AR(1) in the
November/December 2021
following way:
wctþ1 ¼ b0 þ b1wpt þ b2rpt þ b3cpt þ b4wct
þ b5rct þ b6cct þ b7pt þ b8gdt þ b9gnt
þ b10tempt þ b11raint þ b12alt
þ b13dbt þ b14lt þ b15cgt þ b16ght
þ b17mtt þ b18rut þ " t:
(1)
In the abovementioned expression, all of the
18 factors are considered along with the historical
consumption of wheat, cotton, and rice. b is
called a hyperparameter whose value is set and
optimized by the algorithm. Where wctþ1 is
wheat consumption for the next year, which can
be replaced by cctþ1 and rctþ1 accordingly, and " t
represents an indefinite error value. The
selected parameters have a positive effect
because of having positive R2 value.
In contrast, (2) represents the AR(1) with
reduced influence factors for wheat. We
obtained a subset of factors in this section for
forecasting wheat consumption that comprised
of wheat consumption (wct), population (pt),
GDP (gdt), and greenhouse gas emission (ght). So
as to build the linear parametric model, we verified
the Pearson correlation coefficient R2 of
subset factors and wheat consumption. Thus, by
including only the selected factors, the AR(1) for
wheat consumption is defined as follows:
wctþ1 ¼ b0 þ b1wct þ b2pt þ b3 lnðgdtÞþ b4ght þ " t:
(2)
The modeling of cotton and rice sample data
is quite similar to that of the wheat, which is
well-briefed above. The subset factors of cotton
consumption are wheat production (wpt), wheat
consumption (wct), cotton consumption (cct),
loans (lt), rural population (rut), and CO2 gas
emission (cgt). While subset shows wheat consumption
(wct), rice consumption (rct), population
(pt), and methane gas emission (mtt)as
most influential factors for forecasting rice consumption.
Thus, AR(1) is tested in two scenarios
for each crop, first, by including all the factors
and second, by including subset of factors.
In order to evaluate the appropriate model
for forecasting consumption, we calculated the
49
IEEE Consumer Electronics Magazine - November/December 2021

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - November/December 2021
