Sustainable Development Forecasting of the Agricultural Sector using

Machine Learning

Olena Vasyl’yeva

, Lidiia Horoshkova

, Denis Morozov

and Olena Trokhymets

Department of International Tourism and Economics,

National University «Zaporizhzhia Polytechnic», 64 Zhukovskoho Street, Zaporizhzhia, Ukraine

Department of Environmental Studies, National University of «Kyiv-Mohyla Academy», Kyiv, Ukraine

Department of National Economy, Marketing and International Economic relations,

Classic Private University, 70b Zhukovskoho Street, Zaporizhzhia, Ukraine

Keywords: Agricultural Sector, Labour Potential, Sustainable Development, Labour Productivity, Artificial Neural

Network.

Abstract: Sustainable development paradigm is a combination of economic, social and environmental components

represented by a significant number of interconnected factors. Their comprehensive impact determines the

ways and dynamics of achieving sustainable development goals. Sustainable development forecasting is

accompanied by the analysis and processing of a significant set of indicators and requires special methods of

data processing. The neural network modelling allowed to form a multifactorial impact model on the final

indicator, namely labour productivity, according to the sustainable development goals. The proposed model

allows not only to model and forecast, based on the previously obtained indicators and their dynamics, but

also to set target benchmarks to obtain a range of possible scenarios of system development, which depends

on the forecasting conditions and parameters. They do not only increase the validity of managerial decision-

making, but also ensures relevant adaptation of the management object to the changing environment, affects

not only the final result, but also the process of its achievement, including optimization of sustainable

development levers.

1 INTRODUCTION

The global goal of sustainable development is

harmonization of economic, social and environmental

trends of mankind`s way of life; targeting general

well-being due to the ecologically balanced and

socially-oriented economy. The main goal of

sustainable development is providing food for the

population of Earth, so the most relevant and

developed sectoral level problem is of sustainable

development for the agro-industrial complex.

Permanent growth of agricultural production,

better rural quality of life and environmental

preservation are the determinants of economic growth

of the national agricultural sector and its sustainable

development. Human capital acts as a determining

https://orcid.org/0000-0003-2859-3592

https://orcid.org/0000-0002-7142-4308

https://orcid.org/0000-0001-9446-8736

https://orcid.org/0000-0001-7587-7948

lever of the national economic growth in general and

the agrarian sector, in particular, by implementing

labour. It causes lower production costs, higher

productivity and profits, leads to accumulation of

production capital, which ensures sustainable

development of the national economy.

Rapid economic development and quality of life

improvement are achieved in close connection with

sustainable development, but require effective

management of natural and technological resources at

both global, regional, national or local levels. New

challenges and sustainable development indicators

are constantly emerging, which require setting the

priority for each problem`s decision-making. Since

these indicators are characterized by uncertainty,

vague vision of new problems and indicators`

Vasyl’yeva, O., Horoshkova, L., Morozov, D. and Trokhymets, O.

Sustainable Development Forecasting of the Agricultural Sector using Machine Learning.

DOI: 10.5220/0011347100003350

In Proceedings of the 5th International Scientiﬁc Congress Society of Ambient Intelligence (ISC SAI 2022) - Sustainable Development and Global Climate Change, pages 187-196

ISBN: 978-989-758-600-2

187

interconnection, it is preferable to analyze them using

forecasting models with hidden information, which

cannot be perceived by means of classical analysis

(Al'mukhamedova, 2021).

The paradigm of sustainable development is

based on the combination of economic, social and

environmental components represented by a

significant number of interconnected factors. Its

comprehensive action determines the ways and

dynamics of achieving sustainable development

goals. Managerial decision-making referring to each

impact factor, taking into account their interaction has

to be accompanied by the analysis, and processing of

a significant set of data requiring special methods of

information processing. Solving this problem is

possible using machine learning and parameters`

assessment, which will be considered in the model.

To determine the levers of influence on the target

indices of sustainable development in the agrarian

sector, one of which is labour productivity in

agriculture, it is necessary to evaluate sustainable

development determinants.

The method of artificial neural network can be

used for research. The following stages characterize

the method (Zaporozhchenko et al., 2019):

- search of data;

- preparation and normalization of data;

- choice of type of neuron network;

- experimental choice of network characteristics;

- experimental choice of parameters;

- obtaining an artificial neuron network for

modeling the labour productivity;

- checking of adequacy of the model;

- adjustment of parameters,

- final network training using learning sampling;

- adaptation of a neural network caused by

changes in weight coefficients reflecting network

interconnection and network configuration

adjustment.

2 METHODOLOGY

Among the currently known models and forecasting

methods are (Bizianov, 2021): multiplicative models,

dynamic linear and nonlinear models, threshold

autoregressive models, Kalman filters, time series,

ARMAX models, nonparametric regression models,

artificial neural networks (ANNs), statistical models,

as well as hybrid models, for example, fuzzy artificial

neural networks (NNNs).

Various types of regressions and the models

generated from them, as well as time series, can be

effectively used in cases where the dependence of the

predicted indicator over time is continuous, has a

smooth character and does not contain jumps and

gaps. In the case of forecasting based on non-periodic

data series, in order to obtain an acceptable accuracy

(at least a few percent), one has to take into account a

significant number of terms of the series or regression

coefficients. In addition, when processing non-

periodic signals, both regression and time series give

adequate results only within the interpolation interval.

Artificial neural networks are more flexible than

the above models, due to the presence in them of a

complete relationship between input and intermediate

variables, as well as the possibility of introducing

non-linearity into the activation functions (Khaykin,

2006). This explains their expanding application in

solving computational, statistical, prognostic and

other problems. The disadvantage of “classical”

ANNs is that for their training it is necessary to have

a sufficiently large amount of initial data, which is not

always possible.

Linear methods are traditionally used for

macroeconomic forecasting. One of their

disadvantages is not taking into account hidden linear

relationships between model input parameters.

Forecasting makes it possible to obtain a set of

possible scenarios for system`s development, which

depends on the conditions and parameters of

forecasting. It causes application of a wide range of

methods, one of which is the method of artificial

neural networks.

Artificial neural networks are quite effective when

solving problems of predicting the behavior of

complex systems and the selection of unknown

parameters that link the characteristics of complex

objects, including economic systems (Romanchukov,

2019).

Parameters characterizing economic, social and

environmental impacts have been chosen as

determinants (controlled parameters) affecting

sustainable development in the agrarian sector.

Labour productivity is considered as final indicator,

which is a factor characterizing the efficiency of

labour potential application. Input indicators for

modelling by the method of artificial neural networks

are the following:

Y - labour productivity, UAH per worker;

- average monthly nominal wage of full-time

workers in agriculture, UAH;

- energy security (power capacities/sown area),

kW/100 hа;

- power-weight ratio (power capacities

/number of employees), kw/per capita;

- number of tractors per 100 hectares of sown

area;

ISC SAI 2022 - V International Scientiﬁc Congress SOCIETY OF AMBIENT INTELLIGENCE

188

- number of tractors per 1,000 employees;

- mineral fertilizers per 1 hectare of sown area,

kg (nutrients);

- organic fertilizers for agricultural crops, per

1 hectare of sown area, tons;

- stationary and mobile sources of air pollution,

per capita, kg;

- total waste accumulated at landfill sites per

capita, kg;

- average index of regional human

development;

- capital investment per capita, UAH.

2.1 Data and Justification

2008-2019s statistical data was used to forecast

sustainable development parameters. Labour

productivity growth in agriculture was considered as

a target value for the resulting indicator according to

the sustainable development goals (relative to 2015)

(Natsional'na dopovid', 2017). An adverse trend

characterizes agriculture in Ukraine: labour

productivity and average monthly nominal wage

growth show ahead of wage rates (Fig. 1), which, in

our opinion, brakes causal relationship (Vasyl'yeva,

2021): The rate of wage growth surpasses labour

productivity growth rate (2008-2012). During 2013-

2014, there was a positive tendency: labour

productivity growth rate was ahead of wages growth

rate. Wages growth rate has been surpassing the

labour productivity growth rate since 2015.

Figure 1: Chain growth rates of labour productivity and

average monthly nominal wages in agriculture.

Thus, the formation of labour remuneration

mechanisms in agriculture in Ukraine almost does not

depend on the economic results, the employees fall

short of labour remuneration in comparison with their

efforts. Low productivity growth rates, in turn, do not

form the background for higher income and better

quality of life (Vasyl'yeva, 2021). Positive

interdependence of productivity and remuneration

allows us to conclude that it is necessary to include

parameter X

(average monthly nominal wages of

full-time workers in agriculture) into the model of

labour potential assessment.

Scientific and technological progress, technical

equipment, and latest technologies require certain

level of personnel`s education and qualification being

the determinants of staff efficiency. These qualitative

characteristics of human resources can be described

both by the integral index of regional human

development (X

) and by the production facilities

and technologies applied by personnel when working

- X

Taking into account the influence of air pollution

from stationary and mobile sources (X

) and total

waste accumulated at landfill sites (X

) on economic

growth fully corresponds to the sustainable

development trend, we consider these parameters

relevant to be included into the model for the

assessment and forecasting of labour potential in the

agrarian sector amid sustainable development.

Labour productivity in the context of new

economy is largely determined by qualitative

characteristics of labour potential: level of formal and

informal education, creativity, and well-being.

Qualitative parameters of labour potential in the

model are described using the average index of

regional human development (X

). The index has 33

indicators reduced to 6 subindexes (according to

individual aspects of human development:

reproduction of population; social position;

comfortable life; welfare; worthwhile work;

education) (Rehional'nyy lyuds'kyy rozvytok, 2018).

Besides, the amount of capital investment affects

human development, being directed to the fixed

assets` reproduction, introduction of technical

progress, construction and reconstruction of social

and cultural institutions (X

Thus, the parameters affecting labour potential in

the agrarian sector amid sustainable development

(characterize economic, social and environmental

impacts) have been substantiated for modelling.

2.2 Forecasting Models

Modelling of sustainable development determinants

in the agricultural sector was carried out using a

sample of 120 values of each indicator. A dynamic

range of selected data represent the model`s input

parameters. The correlation matrix of input

parameters shows that the strongest correlation with

other parameters occurs for X

(average monthly

100

110

120

130

140

150

160

year

labour productivity wage

Sustainable Development Forecasting of the Agricultural Sector using Machine Learning

189

nominal wages of full-time employees in agriculture),

(power-weight ratio), X

(mineral fertilizers), X

(average regional human development index), and

(capital investment) (Fig. 2).

Figure 2: The correlation matrix of input parameters of

sustainable development.

The correlation confirms the relationship between

the selected parameters: higher power-weight ratio,

bigger number of tractors and growth of soil

mineralization increase gross output and labour

productivity, which are the basis for higher wages; the

growth of the human development index, capital

investment in education, and health care increase the

level of labour potential and its productivity. The

parameters X

(number of tractors) and X

(stationary

and mobile sources of air pollution) have a negative

impact on labour productivity. X

factor (total waste

accumulated at landfill) does not affect the modelling

result.

Thus, correlation analysis of statistical data has

revealed the impact strength and the relationship

reliability between the model`s factor variables and

the final indicator of labour potential application, i.e.

labour productivity.

2.2.1 Generalized Linear Model (GLM)

Generalized Linear Model is a universal method of

building regression models, which allows to take into

account factors` interaction, the type of distribution

of a dependent variable and assumptions about the

nature of regression dependence. GLM is a well-

designed and easy-to-understand way to build

models.

GLM has the next advantages when doing

analysis in comparison with traditional methods:

- the ability to take into account complex types of

factors` interaction;

- a wide choice of dependence function`s type;

- lack of requirements for the normality of the

response variable`s distribution;

- statistical measurement of various factors`

impact on the observed value;

- obtaining information on the reliability of the

constructed model results.

2.2.2 Artificial Neural Network

Neural network algorithms and technologies as the

latest modelling and forecasting methods applied for

various economic processes have been actively

developed (Al'mukhamedova, 2021; Dawes, 2022;

Maehashi, 2020; Zaporozhchenko et al., 2019).

During network operation, the values of input

variables are put to input elements. Then neurons of

the intermediate and output layers start operating.

Each of them assesses its value of activation,

subtracting from the previous layer`s sum its

threshold. Further step is to develop the activation

function of the presented data, resulting in a neuron

output signal. After performing all neurons`

operations, the output value of the last neurons` layer

is taken for the output value of the entire network.

The system, which could be taught about

significant volumes of information, building

correlations and functional dependencies, which

cannot be detected when using other information

processing methods, is one of the advantages of

system`s forecasting based on artificial neural

networks (ANN).

Neural network modelling benefit is neural

networks` potential to find out optimal indicators for

the tool and build optimal prediction strategy for the

range. Moreover, these strategies can be adaptive,

changing with external factors shift, which is

especially important for the systemic phenomena of

sustainable development.

2.2.3 Random Forest Algorithm

Modern ensemble methods of machine learning for

the regression classification include the Random

Forest method, which is to build an array ("forest") of

decision-making trees, making an average forecast

(regression) of the built trees. Random forest is a

managed learning algorithm. Built "Forest" is an

ensemble of decision trees, which is typically taught

by the method of "bags". The general idea of the

“bags” method is to combine learning models to

increase the overall result.

ISC SAI 2022 - V International Scientiﬁc Congress SOCIETY OF AMBIENT INTELLIGENCE

190

3 RESULTS AND DISCUSSION

To check the statistical significance of the difference

between model`s mean values of input parameters

and to assess the probability of their interaction, the

dispersion analysis of sustainable development

parameters was carried out (Table 1).

The dispersion analysis revealed that parameters

(average nominal wage of full-time employees in

agriculture), X

(number of tractors per 100 hectares

of sown area), X

(mineral fertilizers) have the

biggest impact on labour productivity. X

(power-

weight ratio) has somewhat smaller influence. X

parameter (energy security) is insignificant.

To build a generalized regression model (GLM),

a linear model coefficients were calculated. Building

of a generalized regression model in the Rstudio

software environment:

Call:

glm(formula = Y ~ ., family =

"gaussian", data = scaled_tr)

Deviance Residuals:

Min 1Q Median 3Q Max

-0.248644 -0.061787 0.002466 0.059341

0.281375.

Coefficients:

Estimate Std. Error t value Pr(>|t|)

(Intercept) 0.25518 0.04391 5.811

6.35e-08***

X1 0.26334 0.10127 2.600 0.0106*

X2 -0.11443 0.49408 -0.232 0.8173

X3 0.25227 0.55407 0.455 0.6498

X4 -0.34562 0.19872 -1.739 0.0849

X5 0.02655 0.50460 0.053 0.9581

X6 0.67904 0.09670 7.022 2.03e-10***

X7 -0.09787 0.07328 -1.336 0.1845

X8 -0.10786 0.09928 -1.086 0.2797

X9 0.21204 0.12418 1.707 0.0906

X10 0.12414 0.06547 1.896 0.0606

X11 -0.25292 0.13467 -1.878 0.0631

Signif. codes: 0‘***’ 0.001‘**’

0.01‘*’ 0.05‘.’ 0.1‘ ’ 1

(Dispersion parameter for gaussian

family taken to be 0.01255949).

The resulting equation of regression to find the

resulting indicator (labour productivity) is presented

as:

Y=0,25518+0,26334X

-0,11443X

+0,25227X

0,34562X

+0,02655X

+0,67904X

-0,09787X

0,10786X

+0,21204X

+0,12414X

-0,25292X

(1)

Visualization of the forecasting results by the

method of the generalized regression model (Fig. 3).

A neural network consisting of three layers, each

of which has seven, three and one direct propagation

neurons was studied. As an optimization algorithm

the method of reverse error distribution, the activation

function of hidden layers` neurons – Sigmoid, output

layer – linear, input initialization of scales – arbitrary,

the loss function as the error sum of squares were

used. Network learning consisted of finding and

determining the weights of neurons (synaptic

weights) that minimize the difference between the

target variable and the outcome of the network

(Derbentsev et al., 2020). To teach the network, a data

set consisting of a set of input parameters and the

desired output values (target value of labour

productivity) was applied.

The study modelling was provided by the

programming language R using the Rstudio software

(free environment for software development with free

input code for programming language R applied for

statistical data processing and graphics) (Hornik,

2015). The weights of each layer`s neurons of the

network after learning were obtained. The graphical

illustration of the obtained neural network in the

Rstudio environment is as follows (Fig. 4).

The model based on the decision tree "Random

Forest"(Xie, 2020, Sinha, 2019), including 500 trees

(combinations of parameter values) was analyzed too:

Call:

randomForest(formula = Y ~ ., data =

scaled_tr, ntree = 500, mtry = 3,

importance = TRUE, proximity = TRUE,

oob.prox = FALSE.

Type of random forest: regression.

Number of trees: 500

No. of variables tried at each split:

Mean of squared residuals: 0.01138498

% Var explained: 77.74.

The Random Forest method (Fig. 5, 6) also proves

significant impact of parameters X

(average monthly

nominal wages), X

(mineral fertilizers), X

(number

of tractors), and X

(capital investment) on the

efficiency of labour potential in the agrarian sector

(labour productivity).

The results of forecasting by the method of

artificial neural networks and the method of Random

forest in comparison with the modelled values are

presented by Fig. 6,7, respectively.

Comparative analysis of artificial neural

networks, Random forest method and generalized

linear regression method for predicting sustainable

development in the agricultural sector prove that each

of these methods can be used (Table 2), but, in our

opinion, the most appropriate is the application of

artificial neural networks` method, as it has a number

of advantages.

Sustainable Development Forecasting of the Agricultural Sector using Machine Learning

191

Table 1: Dispersion analysis of sustainable development parameters.

Df Sum Sq Mean Sq F value Pr (>F)

1 2,42E+11 2,42E+11 282,8 < 2Е-16 ***

1 5,75E+09 5,75E+09 6,731 0,0107 *

1 6,93E+09 6,93E+09 8,11 0,0052 **

1 1,13E+10 1,13E+10 13,251 0,0004 ***

1 2,56E+09 2,56E+09 2,994 0,0864 .

1 4,70E+10 4,70E+10 54,978 2,90E-11 ***

1 2,84E+09 2,84E+09 3,322 0,0711 .

1 3,15E+08 3,15E+08 0,369 0,545

1 1,44E+09 1,44E+09 1,683 0,1973

1 2,45E+09 2,45E+09 2,861 0,0936 .

1 3,02E+09 3,02E+09 3,527 0,0630 .

Residuals 108 9,23E+10 8,55E+08

Figure 3: Calculated (black) and forecasted (blue) values of Y (GLM method).

A specific feature of the methodological approach to

building a forecasting model is artificial neural

networks` method, which allows to take into account

a significant number of impact factors and to ensure

minimal forecasting error (Kernasyuk, 2017), the

nonlinearity and interaction of parameters (Maehashi,

2020). Analysis of the obtained models showed that

the forecasting results based on the GLM method give

the closest to the modelled values. Nevertheless, there

are lower values of the RMSE metric (total error of

the predicted value and known value) when teaching

the model. This is because modelling of the most

parameters` values for 2030 was based on the linear

models of approximations. Thus, the GLM model was

tracking the linear patterns of parameters behavior

during training.

Table 2: Forecasting results by different methods.

Metrics

Forecasting methods

Neural

net

Random

forest

GLM

RMSE 24424,36 27835,16 27735,43

0,82862 0,77741 0,77901

Predict for

2030

465768 403835 483210

Modelled

decision

508356 508356 508356

Forecast

accuracy, %

91,62 79,44 95,05

ISC SAI 2022 - V International Scientiﬁc Congress SOCIETY OF AMBIENT INTELLIGENCE

192

Figure 4: Adaptive neural model for forecasting sustainable development in the agrarian sector.

Figure 5: Each predictor`s importance.

Sustainable Development Forecasting of the Agricultural Sector using Machine Learning

193

Figure 6: Calculated (black) and forecasted (red) values of Y (Random forest method).

Figure 7: Calculated (black) and forecasted (green) values of Y (Artificial neural networks method).

However, the results based on the neural network

give the lowest values of the RMSE metric. It means

that the model studied as much as possible real hidden

patterns of the analyzed data (necessary reliable

information for forecasting), and was able to build a

more reliable forecast. This is also proved by the

coefficient of determination R

of the neural network

model. It shows the degree of dispersion, being the

highest in the neural network.

The results based on the Random forest model

give much lower indices than in other models,

therefore, its application in this type of data for this

problem statement, in our opinion, cannot be

considered appropriate.

When doing research, the emulated data obtained

from the training data was applied for testing. Testing

results were similar to the results based on the training

data. In the future, for the accuracy of forecasting, it

is interesting to perform tests based on real historical

data. Their collection is somewhat complicated by the

changed methodology and reporting documentation

of the State Statistics Service of Ukraine.

Thus, based on the results, we can conclude that

the built neural network model gives more reliable

results for forecasting sustainable development

parameters in agriculture and can be used to develop

strategic management trends for labour potential in

agriculture, which will ensure its future development.

4 CONCLUSIONS

The neural network modelling allows to form a

multifactor impact model on the resulting indicator,

namely labour productivity in accordance with

sustainable development goals. The following impact

factors have been identified in the model: the average

monthly nominal wages of full-time employees in

agriculture, UAH (X

); energy security (power

capacities/sown area), kW/100 hа (X

); power-weight

ISC SAI 2022 - V International Scientiﬁc Congress SOCIETY OF AMBIENT INTELLIGENCE

194

ratio (power capacities /number of employees),

kw/per capita (X

); number of tractors per 100

hectares of sown area (X

); number of tractors per

1,000 employees (X

); mineral fertilizers per 1

hectare of sown area, kg (nutrients) (X

); organic

fertilizers for agricultural crops, per 1 hectare of sown

area, tons (X

); stationary and mobile sources of air

pollution, per capita, kg (X

); total waste accumulated

at landfill sites per capita, kg (X

); average index of

regional human development (X

); capital

investment per capita, UAH (X

). X

(average

monthly nominal wage), X

(mineral fertilizers), X

(number of tractors), and X

(capital investment)

have the most significant impact on the result. The

proposed model allows modelling and forecasting,

based not only on previously obtained indicators and

their change dynamics (it is the studied period from

2008 to 2019), but to set targets, which is important

in the context of sustainable development. That is

why there is possibility to have administrative impact

not only on the final result, but also on the process of

achieving it, including optimization. In addition, the

modelling allows to adjust impact factors, if they are

either insignificant, as it has been found out when

modelling, or lose significance due to technological

changes (e.g. energy security and power-weight

ratio). Thus, because of modelling aimed at

forecasting the level of labour potential in the context

of sustainable development, an approach to complex

systems has been used. According to it, each of its

components (impact factors on the resulting

indicator) is also a systemic phenomenon. Modelling

each factor`s behavior allows to affect their dynamics

and effectiveness.

The advantages of the applied neural network

modelling include the fact that there is no need to

check (as in traditional modelling) multicollinearity,

i.e. the linear relationship between factors. In the case

they are detected, the factors are being eliminated. It

devaluates the forecast. Therefore, the applied model

takes into account all input parameters, based on their

practical impact on the final result.

Thus, because of neural network modelling it is

possible to identify strategic trends of labour potential

management in the agricultural sector, as well as

economic, social and environmental activities aimed

at improving the quantitative and qualitative

indicators of human capital.

The proposed model allows not only modelling

and forecasting based on previously obtained

indicators and the dynamics of their change, but also

to set targets to obtain a range of possible scenarios

for system development, depending on forecasting

conditions and parameters, which not only increases

the validity of managerial decision-making. It also

ensures the relevance of management object`s

adaptation to the ever-changing environment;

managerial influence not only on the final result, but

also on the process of its achievement, including the

impact aimed at levers` of sustainable development

optimization.

In further research when determining the strategic

directions of labour potential management, it is

advisable to use other models` parameters to

characterize socio-environmental and economic

aspects, considering their significant effect on the

achievement of sustainable development goals in the

agricultural sector.

REFERENCES

Al'mukhamedova, O. (2021). Primenenie neyrosetevykh

sistem iskusstvennogo intellekta v dostizhenii

ustoychivogo razvitiya turizma. Servis v Rossii i za

rubezhom, 15 (3), 7-17. https://doi.org/10.24412/1995-

042X-2021-3-7-17

Bizianov, E., Gutnik, A., Pogorelov, R. (2021). Fuzzy

artificial neural network without rules for forecasting

and control tasks. Bulletin of DonNU. Series G:

Engineering Sciences. 1, 78-85.

Dawes, J.H.P. (2022). SDG interlinkage networks:

Analysis, robustness, sensitivities, and hierarchies,

World Development, 149, 105693,

https://doi.org/10.1016/j.worlddev.2021.105693

Derbentsev, V., Matviychuk, A., Datsenko, N.,

Bezkorovainyi, V. and Azaryan, A. (2020) Machine

learning approaches for financial time series

forecasting. Proceedings of the Selected Papers of the

Special Edition of International Conference on

Monitoring, Modeling & Management of Emergent

Economy (M3E2-MLPEED 2020) Odessa, Ukraine,

July 13-18, 2020. 434-450. http://ceur-ws.org/Vol-

2713/

Hornik, K. (2015). What is R? “R FAQ” The

Comprehensive R Archive Network.

Kernasyuk, Yu. (2017). Neyronni shtuchni merezhi yak

efektyvnyy instrument adaptyvnoho prohnozuvannya v

ahrarnomu sektori ekonomiky. Naukovi pratsi

Kirovohradsbkoho natsional'noho tekhnichnoho

universytetu. Ekonomichni nauky, 32, 224-231.

Khaykin, C. (2006). Neyronnye seti : Polnyy kurs. Moskva

: Vil'yams,. 1104.

Maehashi, K. & Shintani, M. (2020). Macroeconomic

Forecasting Using Factor Models and Machine

Learning: An Application to Japan. Journal of the

Japanese and International Economies, 101104.

https://doi.org/10.1016/j.jjie.2020.101104

Natsional'na dopovid' «Tsili Staloho Rozvytku: Ukrayina»

(2017).

Rehional'nyy lyuds'kyy rozvytok: stat. byuleten' (2018).

Kyyiv : Derzhavna sluzhba statystyky Ukrayiny.

Sustainable Development Forecasting of the Agricultural Sector using Machine Learning

195

Romanchukov, S., Berestneva, O., Petrova, L. (2019).

Teaching a neural network modeling socio-economic

development of the region. Digital sociology, 2 (2), 34-

40. https://doi.org/10.26425/2658-347X-2019-2-34-40

Sinha, P., Gaughan, A., Stevens, F., Nieves, J., Sorichetta,

A., & Tatem, A. (2019). Assessing the spatial

sensitivity of a random forest model: Application in

gridded population modeling. Computers, Environment

and Urban System, 75, 132-145.

https://doi.org/10.1016/j.compenvurbsys.2019.01.006

Vasyl'yeva, O. (2021). Trudovyy potentsial ahrarnoyi sfery

yak bazovyy imperatyv staloho rozvytku. Zaporizhzhya

: FOP Mokshanov, V.

Xie, X., Wu, T., Zhu, M., Jiang, G., Xu, Y., Wang, X., &

Pu, L. (2021). Comparison of random forest and

multiple linear regression models for estimation of soil

extracellular enzyme activities in agricultural reclaimed

coastal saline land. Ecological Indicators, 120, 106925.

https://doi.org/10.1016/j.ecolind.2020.106925

Zaporozhchenko, V. Y., Shepel, A. V., & Tkachuk, A. V.

(2019). Creation of neuron network productivity of

lucerne in Steppe zone of Ukraine. Agrology, 2(1), 47‒

50. https://doi.org/10.32819/2617-6106.2018.14017

ISC SAI 2022 - V International Scientiﬁc Congress SOCIETY OF AMBIENT INTELLIGENCE

196