







































M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

22 

 

 

 

Article 

Managing risk and volatility in oil-dependent 

economies: the role of advanced predictive 

analytics 
Mahmood Abdoos, Amirali Saifoddin*, Hossein Yousefi, Sattar Zavvari, Ali Majnoon 

School of Energy Engineering and Sustainable Resources, College of Interdisciplinary Science and Technology, University 

of Tehran, Tehran, Iran 

A R T I C L E   I N F O 
 

Article history: 
Received 15 July 2025  
Received in revised form 
20 August 2025 
Accepted 05 September 2025 
 
Keywords: 
Oil price forecasting, Neural networks,  
Economic policy, Risk management, 
Investment strategies  
 
*Corresponding author 
Email address:  
saifoddin@ut.ac.ir 
   
 
DOI: 10.55670/fpll.fuen.4.4.3 

A B S T R A C T 
 

The forecasting of oil production, demand, and prices holds critical significance 
for global economic stability and growth. Oil plays a crucial role in determining 
economic performance, making reliable price estimations essential for shaping 
public policy and guiding investment decisions. In this study, advanced neural 
network models were employed to enhance the accuracy of oil market 
forecasts, with a particular focus on their economic implications. Using Python-
based implementations of Long Short-Term Memory (LSTM), Radial Basis 
Function (RBF), and multilayer perceptron (MLP) networks, the research 
compares the effectiveness of these approaches in crude oil price forecasting. 
The evaluation of model outputs using technical indicators revealed that the 
multilayer perceptron network yielded the best results. During training, it 
reached an average squared error of 55.28, a root mean squared error of 7.43, 
and a mean absolute error of 5.55; while in testing, the values were 116.01, 
12.96, and 10.73, respectively. Overall, the comparative analysis indicates that 
the multilayer perceptron consistently surpassed both LSTM and RBF models 
in minimizing prediction errors. The economic relevance of these findings is 
underscored by the model's potential to enhance decision-making processes for 
investors, policymakers, and oil producers by offering more reliable forecasts. 
By improving accuracy by 20 to 30 percent compared to previous studies, this 
research provides valuable insights into optimizing resource allocation and 
mitigating the economic risks associated with oil price volatility. 
 

 
1. Introduction  

The production of oil is undoubtedly a critical input. All 
countries that produce and export oil are affected by changes 
in their market indicators, including price fluctuations and 
instability. Additionally, crude oil market shocks can have a 
profound economic impact. In this regard, an increase in 
crude oil prices, for example, results in a decrease in energy 
demand and, consequently, a decrease in capital productivity. 
An event that will increase unemployment, assuming nominal 
wages remain stable. Oil shocks have numerous and 
widespread consequences. Several studies on oil prices and 
demand are cited in the background of the research. 
Furthermore, the oil market has always been a volatile 
market characterized by unpredictable events. An 
examination of the recent changes in oil prices and several 
major oil shocks proves this fact. This not only motivates 
researchers to conduct research in this field but has also 
served as a platform for innovative efforts and new models, 
as evidenced by its substantial research output. Economic 

enterprises and governments have a great deal of interest in 
accurately forecasting oil market indicators [1]. Oil price 
changes, like other assets, are based on efficient markets. For 
this reason, the price for the previous period will be 
predicted. The changes can estimate price fluctuations in the 
following periods. For this reason, forecasting takes a 
significant share of oil financial market studies. In most 
countries, the stability of crude oil prices plays an essential 
role in their national security and economic development. 
Research on predicting petroleum prices using neural 
networks has led to the development of several 
methodologies and strategies for analyzing different crude oil 
price data forms. Research has shown the significance of 
crude oil pricing and its influence on corporate operations. 
International trade, global commerce, and macroeconomic 
policy [2]. Additionally, artificial intelligence techniques are 
being applied to many research projects. Researchers are 
continually exploring neural network-based approaches to 
forecasting crude oil prices. The two main directions are as 

 

 

Future Energy 

Open Access Journal 

https://doi.org/10.55670/fpll.fuen.4.4.3 

 

 

 

November 2025| Volume 04 | Issue 04 | Pages 22-30 

Journal homepage: https://fupubco.com/fuen 

 
ISSN 2832-0328 

mailto:saifoddin@ut.ac.ir
https://doi.org/10.55670/fpll.fuen.4.4.3
https://fupubco.com/fuen


M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

23 

 

follows. The first direction is to develop forecasting tools that 
utilize neural networks as a standalone tool or in combination 
with other tools. Without considering the uncertainty in the 
price of crude oil, theoretical development and practical 
implementation will suffer detrimental effects, which is why 
forecasting is necessary. As a result, we anticipate that crude 
oil prices will become increasingly challenging. Neural 
networks are utilized to investigate the correlation between 
crude oil price fluctuations and predict trends in critical crude 
oil markets [3]. 

In this article, an improved deep neural network is 
employed to achieve a more accurate prediction of oil prices 
and demand, as numerous factors, including global supply 
and demand, economic and political developments, climate 
change, and others, influence the price and demand of oil. 
Better results are achieved due to the use of deep neural 
networks. This article utilizes a developed deep neural 
network to analyze oil prices and demand. This article fills in 
the gaps in previous articles. As a result, it is possible to design 
a deep neural network model with layers appropriate for each 
feature or characteristic of the data. It is possible to model 
price and demand using two separate neural networks or a 
hybrid model. Moreover, the model must be optimized by 
changing the network architecture, adding new layers, or 
updating parameters. One of the goals of this research is to 
improve the performance of these three methods compared 
to previous research. In addition to comparing them with 
each other to select the best prediction model, this study also 
compares them with previous studies mentioned in the 
introduction and literature review sections for the single 
selected method. Crude oil is a vital component in the 
manufacturing industry. Crude oil prices fluctuate according 
to the economic principles of supply and demand, making 
them challenging to predict accurately.  Zhang et al. [4] 
suggested that an additional reason for using oil price 
fluctuations is the reaction of oil-related companies along the 
economic value chain. In contrast, when oil prices continue to 
decline, the appetite for oil refineries and chemical companies 
will decrease, resulting in lower operating profits. Companies 
that can accurately predict oil price fluctuations will be able 
to reduce cost risk and maintain sustainable growth. Based on 
Zhang et al. [5] findings, oil price fluctuations significantly 
impact both the real economy and the virtual economy of 
crude oil. These factors impact the export and import sectors 
of oil-producing and exporting nations. As a result, it is 
appropriate to focus on crude oil price forecasting. 
Forecasting models can be classified into classical 
econometric and machine learning approaches [6]. Wei et al. 
[7] employed ARIMA stochastic implementation models and 
GARCH conditional heterogeneity models to predict crude oil 
prices. GARCH-type models were used to capture volatility, 
and crack spread futures outperformed mass random walk 
models. As a result, it has been found that nonlinear GARCH 
models outperform linear models in accurately reflecting 
both long-term memory and volatility asymmetry in prices, 
more so than linear models [8]. 

Lin et al. [9] reported in their study that, in addition to 
classical econometric approaches, machine-learning methods 
have also been employed in recent years to forecast crude oil 
prices. Among these methods, neural networks [10] and 
support vector machines (SVMs) [11] have been used most 
often to simulate the complex characteristics of oil prices. Yu 
et al. [12] propose that all methods and external and internal 
factors in oil prices are aimed at achieving short-term effects 
to maximize performance. Deep learning models such as CNN 
[13] neural networks, deep belief networks (DBNs) [14], and 

long short-term memory (LSTMs) [15] may be helpful for 
investors and market analysts dealing with increasingly 
complex data. Using more layers in these models is intended 
to help investors make informed investment decisions. These 
components are used in neural networks [16]. 

Crude oil price forecasting aims to predict price 
movement one step ahead. To accomplish this, a modeled 
relationship is used in conjunction with data and crude oil 
prices. CNN is considered one of the most effective methods 
[17]. A combination of LSTM and RNN models is 
recommended for long-term forecasting of oil prices and 
demand [18]. LSTM is most important for long-term storage. 
Based on historical price data, Bousari et al. [19] reported that 
many studies employ the ANN approach to predict crude oil 
prices using time series analysis. Additionally, according to 
research, a short-term memory neural network (LSTM) is 
applied to sequential data to learn the pattern of past price 
fluctuations and predict the future. Different approaches have 
been attempted to generate and utilize neural networks in 
studies on predicting crude oil prices. Wang et al. developed a 
model that examines variations in data networks. Crude oil 
price forecasting using time series data involves artificial 
intelligence methods, such as backpropagation neural 
networks, radial basis function neural networks, and learning 
machines. Crude oil price volatility remains influenced by 
sentiment data from news sources, particularly in the short 
term. The price of crude oil can be impacted directly and 
indirectly by emotional data [20]. According to Zhang et al. 
[21], factors such as the Persian Gulf War or Russia's attack 
on Ukraine significantly impact oil prices. Due to the COVID-
19 outbreak, the crude oil market has also experienced short-
term volatility [22]. Studies have shown that news is a crucial 
source of information for gauging market sentiment [23]. The 
accuracy of forecasting short-term stock returns can also be 
improved, which is extremely important for businesses [24]. 
As crude oil prices fluctuate daily, forecasts based on long-
term historical data may not be accurate. 

2. Mathematical equations 

2.1 Criteria for evaluation 
To evaluate the performance of RMSE mean root errors 

and MEA mean absolute errors, considering the evaluation 
criteria, which are used for the accuracy of the evaluation in 
question, the following equations are used: 

RMSE = √
1

N
∗ ∑ (Y1t

iN
I=1 − Yt

i)            (1) 

MAE = √
1

N
∑ |Y1t

i − Yt
i|N

i=1            (2) 

where Y1 is the predictive value, Y is the actual value, and N 
is the number of test samples. 

2.2 Multilayer perception equations 
The neural network changes the connection weight after 

processing each piece of data, based on the amount of error in 
the output compared to the expected result. This example is 
done by monitoring and through backtracking and 
generalizing the least squares algorithm in linear perception. 
Error in the output node, we show n data points. Node values 
are adjusted based on corrections that minimize the amount 
of error in the total output and are: 

𝜀(𝑛) =
1

2
∑ 𝑒𝑗

2
𝑗 (𝑛)             (3) 

Using the gradient, the change in weight is: 



M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

24 

 

∆𝜔𝑗𝑖(𝑛) = −𝜂
𝜕𝜀(𝑛)

𝜕𝜗𝑗(𝑛)
 𝑦𝑖(𝑛)              (4) 

Where in yi is the output of the previous neuron and ƞ is the 
learning rate chosen to ensure that the weights quickly 
converge to the oscillation-free response. The calculated 
derivative depends on the induced local field. It is ʋi that 
changes itself. It is easy to prove that for the output node, this 
derivation can be simplified. 

−𝜕𝜀(𝑛)

𝜕𝜗𝑗(𝑛)
= 𝑒𝑗(𝑛)∅~ (𝜗𝑗(𝑛))                 (5) 

Where ϕ is the derivative of the activation function described 
above, and does not change itself. The analysis for changing 
the weights to a hidden node is more difficult, but it can be 
shown that the corresponding derivative is: 

−
𝜕𝜀(𝑛)

𝜕𝜗𝑗(𝑛)
= ∅~ (𝜗𝑗(𝑛)) ∑ −

𝜕𝜀(𝑛)

𝜕𝜗𝐾(𝑛)
𝜔𝑘𝑗𝐾 (𝑛)          (6) 

It depends on the change in the weights of the k nodes that 
represent the output layer; Therefore, to change the weights 
of the hidden layer, the output layer changes according to the 
derivative of the activation function, and thus this algorithm 
represents a function of the activation function. 

3. Methodology 

3.1 Software 
An improved deep-learning model was used in this study 

to forecast production, demand, and price. Deep learning 
requires significant computing power. There is a parallel 
architecture available in high-performance GPUs that makes 
them suitable for deep learning. Deep learning models are 
often called deep neural networks because they utilize neural 
network architecture. Deep neural networks have a large 
number of hidden layers. In deep learning, data sets are 
labeled, and neural network architectures are used to learn 
features directly from the data without manually extracting 
them. Data characteristics include the collection and 
processing of oil market data (Table 1). In addition to oil price 
information, demand and supply information, global events, 
and other related information are included. Moreover, model 
training is conducted to develop a model that can forecast 
price and demand based on training data.  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Adjusting parameters, applying improved techniques, 
and evaluating model accuracy are all part of this process. 
Several metrics, such as mean square error, assess the 
model's accuracy on test data over different periods. 
Additionally, it includes criteria for predicting, analyzing, and 
modeling. The program also offers optimization and analysis 
of model prediction results and solutions. Optimizing the 
model for higher accuracy in future predictions is also very 
effective. The project aims to predict the final price of oil using 
the prices and demand of oil in different parts of the world, 
and to analyze the developments, which include three key 
components: data preprocessing, modeling, and model 
evaluation. 

3.2 Data preprocessing 
The primary purpose of data preparation is to arrange 

the data for the following modeling step. Data preprocessing 
is a crucial component of machine learning, involving the 
application of mathematical and logical filters to refine the 
data. To achieve this, the following steps will be implemented 
sequentially for data cleaning: 
• Eliminate columns with more than 15 missing values. In 

this step, columns with several missing values exceeding 
ten percent of the total dataset (approximately 15 
instances) will be removed. 

• Exclude the Year column. As the Year feature consists of 
unique values, it does not contribute significantly to model 
training. Consequently, it will be excluded from the dataset 
due to its lack of relevance. 

By following these steps, the dataset will undergo effective 
cleanup, enhancing its suitability for subsequent modeling 
and analysis. 

3.3 Standardizing data  
The scaling of data varies across different columns and 

can have distinct impacts on the learning of data models. To 
address this, we employ the method of min-max 
standardization, which rescales the entire dataset to a range 
of 0 to 1. Additionally, we undertake the following steps for 
data refinement:  
• Remove columns with a correlation exceeding 90%.  
• Exclude data that contains null values. 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  

Table 1. Variables impact the supply and demand of crude oil 

Technical Factors 
 

Economical Factors Political Factors Factors Influencing 
the Market 

Environmental Factors 

  
Crude Oil Quality 

 

 
Cost of Equipment 

 

 
Political Struggles 

 

 
Transaction Volume 

 

Pressure on communities 
to use renewable energy 

 
New Manufacturers 

 
Economic Recession 

 

 
Socio-political conditions of 
oil extractor countries 

 

 
The pricing of energy 
hubs 

 
Temperature changes 
and seasons 

 
Crude oil 
transportation 

 
Gold Price 

 
 

 
Oil Crisis 

 

   
Population growth 

Completion of green 
fields and production 
from brownfields 

 
The amount of oil 
extraction in the world 

  
Enhancing the value of 
the dollar 

 
Global Crisis 

 
OPEC production risk 
and decisions 

 
Focus on renewable 
energy sources 

 
Oil Storage and 
Products 

 

 
Geopolitical 
occurrences and oil 
market shock 

Politics Of Oil Companies  
Non-alignment with 
OPEC Plus 

 
Geographical Disasters 

Innovation and 
Technology 

Economic Development - Supply a-nd Dem-and- 

 
- 

 

 



M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

25 

 

In our predictive analysis, we utilize three distinct methods, 
which we shall briefly introduce. The objective of neural 
networks is to emulate the patterns generated by the human 
brain. Neural networks operate by generating an output 
pattern based on the provided input pattern. Comprising 
multiple processing elements called artificial neurons, neural 
networks receive and process data within the neurons, 
ultimately generating an output. 

3.4 Introduction to the LSTM method (long short-term) 
Long Short-Term Memory (LSTM) networks are an 

improved iteration of recursive neural networks that are 
designed to effectively retain past data within memory. The 
choice of utilizing the LSTM method for predicting oil prices 
and supply-demand dynamics aims to address the inherent 
issue of vanishing gradients common in recurrent neural 
networks. By overcoming this challenge, LSTM enhances the 
quality of forecasting models for price and demand analysis. 

3.5 Introduction to the RBF method 
The RBF (Radial Basis Function) method serves as the 

secondary network employed for predicting oil prices and 
demand. Esteemed for its application in time series 
prediction, this method is commonly employed by industry 
analysts within oil prediction models. The Radial Basis 
Function (RBF) neural network is designed as a feed-forward 
model in which radial basis functions are applied as the 
activation mechanism. Its structure consists of an input layer, 
one or more hidden layers, and an output layer. Owing to this 
architecture, RBF networks are widely recognized as effective 
tools for handling forecasting tasks. 

3.6 Introduction to the multilayer perception (MLP) 
method 
The multilayer perceptron model represents a 

prominent category of neural networks, characterized by 
interconnected layers of neurons. Unlike other deep learning 
algorithms, such as recurrent neural networks, the MLP 
operates solely in a unidirectional manner, transmitting data 
only in a forward direction across the network. MLPs, 
consisting of several elements such as the input, output, and 
hidden layers, are widely acknowledged as the most often 
used architecture in neural networks. Their applicability in 
forecasting oil prices and demand is highly regarded within 
the field. 

4. Results and discussion 

Following the data preprocessing stage, 149 instances 
and 120 columns remain. In this section, two models, namely 
LSTM and RBF, were utilized. The results of the error and the 
accuracy of each model were presented separately. To 
evaluate the performance, the dataset was divided into two 
subsets, namely training and testing, with 70% used for 
training and 30% reserved for testing. Before splitting, to 
avoid selection bias, the data was randomly shuffled. Each 
model was then trained for a total of 300 iterations.  

4.1 LSTM model (long short-term memory) 
The Long Short-Term Memory (LSTM) model shown in 

Figure 1 follows a sequential neural network structure, with 
an LSTM layer with 60 neurons as the main layer. The neurons 
of the LSTM layer capture temporal dependencies and learn 
patterns in time series data, such as changes in oil prices. The 
first layer is an input layer that encapsulates past oil price 
data; this data is then analyzed by the LSTM layer. At this time, 
the information across different intervals is remembered by 
the 60 neurons of the LSTM layer. The LSTM layer specifically 
adapts to work with challenges, like vanishing and exploding 

gradients, usually raised in conventional recurrent neural 
networks (RNNs). This underlying layer is followed by other 
layers that fine-tune the extracted features, ultimately 
sending the information to the output layer, which provides 
accurate price predictions. The diagram in Figure 1 illustrates 
how data flows through these layers and through the 60 
neurons in the LSTM layer; the 60 neurons express long-term 
dependencies. The Figure shows how the model converts the 
input data into trusted predictions, additionally helping the 
decision-making process, in the economic and energy sectors, 
more specifically. According to Figure 2, the Long Short-Term 
Memory (LSTM) method demonstrates a training loss of 
509.70 and a testing loss of 529.98, placing it in the middle 
performance range compared to the other two selected 
methods. While this indicates that the LSTM model performs 
reasonably well, it neither outperforms nor underperforms 
the best and worst methods evaluated in the study. This level 
of loss suggests that the LSTM model successfully captures 
key patterns in the data during training but exhibits some 
level of prediction error during testing, which is characteristic 
of time-series forecasting challenges. Compared to the other 
models, the LSTM strikes a balance between training and 
testing accuracy, making it a viable option for oil price 
forecasting, although further tuning or alternative methods 
may provide additional improvements. Figure 2 visually 
highlights the LSTM's relative position, showing how it 
balances accuracy and loss compared to the other models (the 
model error diagram). 

 
Figure 1. Python code model ™ for price and demand forecasts 

Figure 2. Error chart using the LSTM method to predict demand and 
oil prices 

4.2 RBF model (radial base function) 
The model consists of two fully connected layers and one 

RBF layer. A view of the model is shown in Figure 3. The 
model error diagram is shown in Figure 4. According to Figure 
4, the Radial Basis Function (RBF) method had the highest 



M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

26 

 

casualty rates of the three models, with RBF training casualty 
rates of 763.51 and RBF testing casualty rates of 836.33. In 
terms of minimizing loss, this indicates it performs poorly and 
is much worse than both the LSTM and MLP methods. The 
increased casualty rates suggest that the RBF model is having 
trouble estimating patterns in the data during the training 
and testing phases. The higher error rates in both 
environments indicate a greater challenge in providing a 
discerning outcome, which highlights the limitations of the 
RBF model for forecasting oil prices. On the contrary, both the 
LSTM model and the MLP models performed significantly 
better based on their reduced casualty rates. Figure 4 visually 
underscores the RBF method’s comparatively poor 
performance, making it less favorable for applications that 
require high accuracy, such as economic and energy market 
forecasting. 

Figure 3. Python code of the RBF model for price and demand 
forecasts 

Figure 4. RBF model error in Python software for predicting 
demand and oil prices 

4.3 Multilayer perceptron model 
Multilayer neural networks, particularly multilayer 

perception networks, are widely acknowledged by 
researchers as powerful approximations. It is believed that 
these networks, provided they possess sufficient layers and 
neurons, can estimate any nonlinear transformation with the 
desired level of accuracy. As such, the multilayer perception 
network stands as one of the most versatile and successful 
prediction models. The multilayer perception neural 
network, also known as MLP, utilizes the post-error learning 
rule. This learning method serves as a generalization of the 
least squares error algorithm, based on the principle of error 
correction learning. The algorithm consists of two essential 
paths: the forward path and the backward path. During the 
forward pass, the input vector moves through the 
intermediate layers to the output layers, producing their 

effects. On the return path, the network parameters are 
adjusted, following the principles of the error correction law. 
Despite the proficiency of the aforementioned models, they 
do not perform optimally on data lacking a recursive 
structure, where the order of properties is significant, such as 
in text data. To address this limitation, a three-layer 
perception with a configuration of 60, 50, and 40 neurons, 
respectively, was employed, yielding the best performance. A 
visual representation of the model is displayed in Figure 5, 
while the error chart is illustrated in Figure 6. 

Figure 5. Python Code for the three-layer perceptron model for price 
and demand forecasts 

Figure 6. Error graph of the 3-layer perceptron model for forecasting 
demand and oil prices  

As depicted in Figure 5, each neuron within these layers 
plays a critical role in transforming the inputs from the 
preceding layer. The inputs are combined using weights 
specific to each neuron, determining the influence of each 
input on the neuron's output. After that, the sum of inputs is 
passed through an activation function, which causes 
nonlinearity in the network. Nonlinearity is crucial for 
enabling the model to learn complex patterns and 
relationships within the data. At this stage, each neuron 
generates an output that is specific to each neuron and 
reflects a new feature or attribute that is the nonlinear 
combination of the inputs. The outputs of the neurons of this 
layer establish a feature vector of the data that is comprised 
of the processed input. The feature vector will be reassigned 
as input to the next layer, and the process of transformation 
will continue. The feature vector will be reassigned as input 
to the next layer, and the process of transformation will 
continue. Figure 5 illustrates the effect of this structure. From 
this Figure, it can be seen that each layer in the network 
transforms the data, developing to deeper levels of 
abstraction and therefore extracting increasingly refined 
features from the initial data, and leading to more reliable 
predictions. Neural networks utilize this hierarchical 



M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

27 

 

structure, which can lead to more reliable information 
management and modeling in complex systems, such as oil 
price fluctuations. In Figure 6, it can be observed that the 
multilayer perceptron (MLP) approach has the lowest loss 
rates among the models under evaluation, with a training loss 
rate of 55.28% and a testing loss rate of 116.01%. This 
indicates that, in terms of minimizing loss, the multilayer 
perceptron model is comparatively the best, and it performed 
significantly better than the LSTM and RBF models. The 
notably low loss rates indicate that the MLP model is highly 
effective in capturing patterns during both the training and 
testing phases, achieving superior predictive accuracy. This 
reduction in loss makes it an ideal choice for applications 
where minimizing prediction error is critical, such as oil price 
forecasting. The MLP method’s ability to consistently deliver 
the lowest casualty rates positions it as a top choice among 
the models studied. Figure 6 visually emphasizes the MLP 
method's clear advantage in the discussion of casualties, 
highlighting its potential for making highly accurate 
predictions and ensuring reliability in decision-making 
processes in volatile markets. 

4.4 Comparison of oil forecast models 
The results of the models are presented in Table 2. The 

average squared error in the training mode for the LSTM, RBF, 
and PERCEPTRON models decreased significantly, reaching 
55.28 (from the initial values of 509 and 763.51, respectively). 
Similarly, in the testing mode, the average squared error was 
reduced to 116.01 (from the initial values of 529.98 and 
836.33) for the respective models. In terms of the root mean 
squared error, the LSTM, RBF, and PERCEPTRON models 
achieved values of 7.43 (initially starting at 22.57) and 12.96 
(from the initial values of 23.02 and 28.91) in the training and 
testing modes, respectively. Regarding the mean absolute 
error, the LSTM, RBF, and PERCEPTRON models obtained 
values of 5.55 (from the initial values of 19.56 and 21.95) and 
10.73 (initially starting at 20.84 and 23.71) in the training and 
testing modes, respectively. 

4.5 Validation 
By performing validation using the LSTM method, it has 

been analyzed that this study performed better than the 
previous studies in the compared parameters, and it can be 
used for more accurate predictions with a lower percentage 
of error (Table 3). The study's results on oil price forecasting 
using neural networks, specifically LSTM, RBF, and multilayer 
perceptron (MLP) models, indicate a significant improvement 
in predictive accuracy compared to previous methodologies.  
Model performance metrics: The multilayer perceptron 
(MLP) model demonstrated superior performance with the 
following metrics: 
Training Phase: 
- Average Squared Error (ASE): 55.28 
- Root Mean Squared Error (RMSE): 7.43 
- Mean Absolute Error (MAE): 5.55 
 
 
 
 
 
 
 
 
 
 
 
 

Testing Phase: 
- Average Squared Error (ASE): 116.01 
- Root Mean Squared Error (RMSE): 12.96 
- Mean Absolute Error (MAE): 10.73 
The study asserts that the MLP model outperformed both 
LSTM and RBF models in terms of error metrics. This 
supports the claim that the MLP method is more effective for 
oil price forecasting, achieving a performance improvement 
of 20-30% over traditional models that utilized only one or 
two methodologies. Data preprocessing involved cleaning 
and standardizing the dataset, which consisted of 149 
instances and 120 columns after the initial data cleaning 
process. The dataset was split into training (70%) and testing 
(30%) sets, ensuring unbiased selection through shuffling. 
Model training: Each model was trained for 300 iterations, 
with the LSTM model configured with 60 neurons in its layers. 
The training process aimed to minimize prediction errors, as 
indicated by the performance metrics. 
The results confirm the success of the multilayer perceptron 
model in predicting oil prices, consistently outperforming 
LSTM and RBF in terms of error rates. The method applied, 
which includes detailed data preprocessing, rigorous model 
training, and evaluation, ensures that the findings presented 
are trustworthy and relevant, meeting the research 
specifications outlined in the article. This thorough process 
not only fosters confidence in the quality of the findings but 
also contributes to the growing discourse on oil price 
prediction techniques. Accurate oil price forecasts are crucial 
in developing global economic policy, risk management, and 
resource allocation as oil remains a critical pillar of the 
economy.  

4.6 Global economic policy 
Oil prices have significant effects on the rate of inflation, 

the value of exchange rates, and the economic activity of what 
are typically oil-importing nations, as well as oil-exporting 
nations. With credible forecasts, countries can:  
Formulate effective monetary and fiscal policies: Central 
banks and finance ministries rely on oil price forecasts to 
adjust interest rates, manage inflation, and design fiscal 
policies that strike a balance between growth and stability. 
For instance, governments may take steps to cushion 
inflationary impacts during periods of forecasted high oil 
prices. 
Stabilize currency and trade balances: For oil-exporting 
countries, a good forecast allows for an assessment of export 
revenues, which, in many cases, constitute a significant 
portion of GDP. Conversely, oil-importing countries can better 
plan their foreign exchange requirements, allowing them to 
control currency volatility. 
Develop energy policies: Governments utilize forecasts to 
adjust different energy subsidies, taxes, and strategic reserve 
levels. Governments forecast and plan for energy security 
while facilitating transitions to renewable energy sources. 

 

  

 

 

 

 

 

 

Table 2. Comparison of errors between LSTM & RBF and perceptron models 

Test Train 
 

Model 

MAE RMSE MSE Loss MAE1 RMSE1 MSE1 Loss  

20.84 23.02 529.98 529.98 19.56 22.57 509.70 509.70 LSTM 

23.71 28.91 836.33 836.33 21.95 27.63 763.51 763.51 RBF 

10.73 12.96 116.01 116.01 5.55 7.43 55.28 55.28 
PERCEPTR

ON 

 



M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

28 

 

Accurate forecasts can enable policymakers to reduce their 
reliance on oil when prices are expected to rise and to make 
more informed decisions about replacing new investments in 
energy with alternative energy sources. 

Table 3. Validation of predictions made with previous studies 

MODEL MSE MAE RMSE 

Proposed 
LSTM 

529.98 20.84 23.02 

Proposed 
ARIMA [4] 

1047.851 28.699 - 

LSTM-ANN 
[25] 

699.98 14.56 - 

 

4.7 Risk management 
Unpredictable oil prices lead to enormous risks for a 

firm, the financial market, and the economy as a whole. 
Reliable forecasting substantially reduces these risks by 
allowing them to use some key strategies: 
Facilitating hedging strategies: Firms that have a lot of 
exposure to oil prices, such as petroleum companies, airlines, 
and energy-dependent firms, can employ accurate forecasts 
to hedge against price movements. These firms protect 
themselves against unforeseen spikes in oil prices or dramatic 
declines by locking in prices based on the previous forecasts 
in the future contracts or similar documents. 
Improving investment planning: Investors in the oil sector 
can utilize forecasting for investment purposes in order to 
analyze the profit potential of long-term projects, including 
exploration of oil reserves and capital projects. Accurate 
forecasting will help companies avoid overinvestment, and 
equally important, under-investment, while aiding in faster 
capital utilization. 
Managing macroeconomic risks: Countries that rely on oil 
revenues can experience sudden growth problems and 
deterioration of public finances due to price shocks. 
Forecasting tools permit governments to set stabilization 
processes into motion (e.g., sovereign wealth funds, counter-
cyclical fiscal policies) to help insulate their economy against 
unexpected shocks. 

4.8 Resource allocation 
Reliable oil price forecasts are essential for government 

and industry sector productivity as they guide the effective 
use of resources, and economic stabilization and growth 
potential. 
Optimizing investment in energy projects: Oil industry 
players use forecasts to make decisions about where and 
when to explore, drill, and produce. Adequate planning based 
on accurate predictions will enable oil and gas companies to 
invest their capital efficiently, preventing production during 
times of low demand and underproduction during periods of 
higher prices. 
Optimizing the energy mix: Forecasts regarding oil prices can 
help public policymakers know how to steer the national 
energy portfolio. Oil prices can have a cyclical effect, where 
higher oil prices may support quantitative investment in 
renewable energy, and lower prices are likely to result in 
increased dependence on fossil fuels. If decisions are based on 
accurate forecasts, public agencies and corporations can 
make informed and productive arguments against 
diversification as a factor of security, and as a cost 
management concern. 
Guiding infrastructure development: Forecasting is critical 
for states, provinces, or countries that rely on oil imports, as 
they must make long-term capital investments in 

infrastructure such as pipelines, refineries, and storage 
facilities. The integrity of forecasting efforts, i.e., a better price 
trend analysis, better identifies when you are overspending 
or underusing facilities based on what was inaccurately 
forecasted. 

4.9 Policy Implications 
Better forecasts, particularly in oil-dependent 

economies, can aid the policy-making process by providing 
greater clarity on price movements now and into the future. 
Being able to project prices accurately enables policymakers 
to create policies that mitigate the potential effects of price 
volatility. By considering and revising their policy 
frameworks and strategies, they can proactively limit the 
impacts of price movements, such as inflationary or abrupt 
cuts resulting from fiscal practices.  Having better forecasts 
provides considerable opportunities for the government to 
enhance fiscal indicators, such as oil production, prices, and 
demand, similar to those from this study, which allows the 
government to adjust fiscal instruments, i.e., 
subsidies/taxation, etc., accordingly in regard to changes in 
oil revenues. If the government starts relying on accurate 
forecasting of price movements, it will be able to prepare well 
in advance to update its fiscal budget and economic plan, 
thereby reducing the possibility of unexpected deficits or 
inflationary policy practices caused by sudden exogenous 
shocks. 

Reliable forecasting also supports the establishment of 
long-term energy trajectories. With price projections, 
policymakers can decide whether to diversify the economy, 
further invest in renewable energy, or expand strategic oil 
reserve inventories. Using one of the more accurate 
forecasting methods described in hinter, namely, neural 
network models, and particularly the multilayer perceptron 
model used in this study, gives governments a better basis for 
these decisions on weighing the advantages and risks of 
continued dependency on oil, with transitioning toward 
alternative energy sources. In summary, reliable oil price 
forecasts not only provide policymakers with the capacity to 
support stability in national economies and improve public 
financial management but also enable a measured approach 
towards policies that balance the viability of short-term 
resilience with long-term sustainability. 

4.10 Management and strategy 
The upgraded forecasting model developed in this 

research offers an excellent opportunity for oil producers and 
investors to enhance decision-making, mitigate uncertainty, 
and hedge against market fluctuations. Several potential 
applications are outlined below:  
• Optimizing investment timing: With the accuracy of the 

multilayer perceptron model in predicting oil prices, 
companies and investors can use the model to better 
predict price movements. This allows them to plan when to 
make major decisions, such as increasing or decreasing 
production, increasing capacity, or investing in a new 
exploration program, so that capital expenditures align 
with prices and yield a stronger return on investment. 

• Hedging and risk management: In the context of hedging, 
accurate price forecasting is the key to crystallizing 
effective hedging strategies. The greater the specificity in 
price forecasted moves, the better firms will be able to 
utilize contracts, options, etc. To protect themselves from 
unpredictable and unforeseen moves, which will preserve 
their profit margin and certainly make their revenue 



M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

29 

 

streams from incredibly volatile markets to some 
predictable extent. 

• Reducing market uncertainty: The level of accuracy 
attached to the neural network model, an improvement of 
20–30% against previous studies, fundamentally improves 
the uncertainty normally attached to oil price volatility. 
More sophisticated oil price forecasting enables firms to 
make less speculative decisions regarding supply chain 
management, long-term contracts, and selling prices. 
Furthermore, greater predictability will improve investor 
confidence in firms' ability to deliver returns over the 
investment horizon they need, making it easier for them to 
choose a stable sector and return, thereby placing them in 
a less volatile risk position. 

• Capital allocation and portfolio diversification: Investors 
can measure the type of exposure they want to oil-related 
assets or sectors and decide if they want to gain exposure 
at any point going forward to other sectors. The investment 
was about helping improve the forecast capabilities so that 
there is more clarity on future market fundamentals and 
conditions around the oil price. This helped with 
optimizing the risk and opportunity balance in their 
portfolios across the investment buckets. In conclusion, as 
the leading user of cost principles, applying new generation 
forecasting models has significantly assisted both the 
company and investors in improving profit generation, 
developing and maintaining risk prudence, and enhancing 
mechanisms for capitalizing on potential trading 
opportunities. 

5. Conclusion 

This research makes a significant contribution to 
economics and management, as it offers a broader and more 
reliable method for forecasting oil pricing and demand using 
deep neural networks. Of the models employed, the three-
layer perceptron exhibited the greatest predictive power, 
which is a significant finding for decision-making in oil-
dependent economies, where oil prices can create significant 
uncertainty and can jeopardize fiscal stability and influential 
investment decisions. This improvement, which identifies 
that the model's accuracy is improved by 20%-30% 
compared to prior models, provides stakeholders with 
opportunities to make the best and most informed capital 
allocation, risk management, and long-term decisions. From 
an economic perspective, employing these models can 
mitigate some of the adverse impacts of price volatility, as 
they enable governments and businesses to implement 
steadier fiscal policies, reduce uncertainty surrounding 
revenues, and allocate resources more effectively. For oil 
companies, such projections can help firms make better 
decisions around production scheduling, capital investment 
timing, and apply them to hedging strategies, which will 
improve profitability and reduce their vulnerability to sudden 
market shocks. This information also enables investors to 
utilize it more effectively in protecting their portfolios and 
refining risk management strategies in energy markets. 
Moreover, the use of advanced neural networks in 
conjunction with forecasting processes, specifically economic 
forecasting, makes analyses of what drives energy markets 
more systematic and transparent for stakeholders, providing 
them with additional valuable data to predict trends and take 
action based on facts rather than speculation. When economic 
forecasting incorporates the use of artificial intelligence, it 
has the potential to be influential in reducing uncertainty and 
facilitating more consistent, sustainable development and 
growth. Ultimately, this research also demonstrates the 

expanding role of intelligent predictive models that futures 
and options possess in addressing global oil market 
challenges, and highlights the need for economists and 
students to consider investing in advanced analytical 
capabilities, both in economic and managerial contexts. 

Ethical issue 
The authors are aware of and comply with best practices in 
publication ethics, specifically concerning authorship 
(avoidance of guest authorship), dual submission, 
manipulation of figures, competing interests, and compliance 
with policies on research ethics. The authors adhere to 
publication requirements that the submitted work is original 
and has not been published elsewhere in any language. 

Data availability statement 
The manuscript contains all the data. However, more data will 

be available upon request from the corresponding author. 

Conflict of interest 

The authors declare no potential conflict of interest. 

References 

[1] H. Miao, S. Ramchander, T. Wang, and D. Yang, 

"Influential factors in crude oil price forecasting," 

Energy Economics, vol. 68, pp. 77-88, 2017. 

[2] Z. A. Sadik, P. M. Date, and G. Mitra, "Forecasting 

crude oil futures prices using global macroeconomic 

news sentiment," IMA Journal of Management 

Mathematics, vol. 31, no. 2, pp. 191-215, 2020. 

[3] M. Wang et al., "A novel hybrid method of forecasting 

crude oil prices using complex network science and 

artificial intelligence algorithms," Applied energy, vol. 

220, pp. 480-495, 2018. 

[4] K. Zhang, M. Hong, K. Zhang, and M. Hong, 

"Forecasting crude oil price using LSTM neural 

networks," Data Sci. Financ. Econ, vol. 2, pp. 163-180, 

2022. 

[5] X. Zhang, C. Zhao, X. Zhou, X. Wu, Y. Li, and M. Wu, 

"Capital market and public health emergencies in 

Chinese sports industry based on a market model," 

Data Science in Finance and Economics, vol. 3, no. 2, 

pp. 112-132, 2023. 

[6] H. Abdollahi, "A novel hybrid model for forecasting 

crude oil price based on time series decomposition," 

Applied energy, vol. 267, p. 115035, 2020. 

[7] Y. Wei, Y. Wang, and D. Huang, "Forecasting crude oil 

market volatility: Further evidence using GARCH-

class models," Energy Economics, vol. 32, no. 6, pp. 

1477-1484, 2010. 

[8] B. Wu, L. Wang, S.-X. Lv, and Y.-R. Zeng, "Forecasting 

oil consumption with attention-based IndRNN 

optimized by adaptive differential evolution," Applied 

Intelligence, vol. 53, no. 5, pp. 5473-5496, 2023. 

[9] Y. Lin, Y. Xiao, and F. Li, "Forecasting crude oil price 

volatility via a HM-EGARCH model," Energy 

Economics, vol. 87, p. 104693, 2020. 

[10] M. M. Mostafa and A. A. El-Masry, "Oil price 

forecasting using gene expression programming and 

artificial neural networks," Economic Modelling, vol. 

54, pp. 40-53, 2016. 

[11] Y. Zhao, J. Li, and L. Yu, "A deep learning ensemble 

approach for crude oil price forecasting," Energy 

Economics, vol. 66, pp. 9-16, 2017. 



M. Abdoos et al. /Future Energy                                                                                             November 2025| Volume 04 | Issue 04| Pages 22-30 

30 

 

[12] L. Yu, X. Zhang, and S. Wang, "Assessing potentiality of 

support vector machine method in crude oil price 

forecasting," EURASIA Journal of Mathematics, 

Science and Technology Education, vol. 13, no. 12, pp. 

7893-7904, 2017. 

[13] K. He, Q. Yang, and Y. Zou, "Crude Oil Price Prediction 

using Embedding Convolutional Neural Network 

Model," Procedia Computer Science, vol. 214, pp. 959-

964, 2022. 

[14] I. Méndez-Jiménez and M. Cárdenas-Montes, "Time 

series decomposition for improving the forecasting 

performance of convolutional neural networks," in 

Advances in Artificial Intelligence: 18th Conference of 

the Spanish Association for Artificial Intelligence, 

CAEPIA 2018, Granada, Spain, October 23–26, 2018, 

Proceedings 18, 2018, pp. 87-97: Springer. 

[15] C. Liu, W. Hou, and D. Liu, "Foreign exchange rates 

forecasting with convolutional neural network," 

Neural Processing Letters, vol. 46, pp. 1095-1119, 

2017. 

[16] J. Zhang, L. Xing, Z. Tan, H. Wang, and K. Wang, "Multi-

head attention fusion networks for multi-modal 

speech emotion recognition," Computers & Industrial 

Engineering, vol. 168, p. 108078, 2022. 

[17] X. Ma, Z. Dai, Z. He, J. Ma, Y. Wang, and Y. Wang, 

"Learning traffic as images: A deep convolutional 

neural network for large-scale transportation 

network speed prediction," Sensors, vol. 17, no. 4, p. 

818, 2017. 

[18] Z. Cen and J. Wang, "Crude oil price prediction model 

with long short term memory deep learning based on 

prior knowledge data transfer," Energy, vol. 169, pp. 

160-171, 2019. 

[19] G. A. Busari and D. H. Lim, "Crude oil price prediction: 

A comparison between AdaBoost-LSTM and 

AdaBoost-GRU for improving forecasting 

performance," Computers & Chemical Engineering, 

vol. 155, p. 107513, 2021. 

[20] Y. Hu, J. Ni, and L. Wen, "A hybrid deep learning 

approach by integrating LSTM-ANN networks with 

GARCH model for copper price volatility prediction," 

Physica A: Statistical Mechanics and its Applications, 

vol. 557, p. 124907, 2020. 

[21] Z. Zhang, M. He, Y. Zhang, and Y. Wang, "Geopolitical 

risk trends and crude oil price predictability," Energy, 

vol. 258, p. 124824, 2022. 

[22] Z. Niu, Y. Liu, W. Gao, and H. Zhang, "The role of 

coronavirus news in the volatility forecasting of crude 

oil futures markets: evidence from China," Resources 

policy, vol. 73, p. 102173, 2021. 

[23] C.-H. Hui, C.-F. Lo, C.-H. Cheung, and A. Wong, "Crude 

oil price dynamics with crash risk under fundamental 

shocks," The North American Journal of Economics 

and Finance, vol. 54, p. 101238, 2020. 

[24] Y. Bai, X. Li, H. Yu, and S. Jia, "Crude oil price 

forecasting incorporating news text," International 

Journal of Forecasting, vol. 38, no. 1, pp. 367-383, 

2022. 

[25] A. L. Zahouani and H. Boubaker, "Forecasting Crude 

Oil Price with Hybrid Approaches," Rev. Econ. Financ., 

vol. 21, pp. 564-576, 2023. 

 

 

 
This article is an open-access article distributed under the 

terms and conditions of the Creative Commons Attribution 

(CC BY) license 

(https://creativecommons.org/licenses/by/4.0/). 

https://creativecommons.org/licenses/by/4.0/

