Proceedings of the Sixth International Conference on Inventive Computation Technologies [ICICT 2021]

IEEE Xplore Part Number: CFP21F70-ART; ISBN: 978-1-7281-8501-9

Forecasting by Machine Learning Techniques and

Econometrics: A Review
G. Shobana 1 K. Umamaheswari2
Assistant Professor Assistant Professor
Department of Computer Applications Department of Computer Science
Madras Christian College Bharathi Women’s College
Chennai Chennai
Email: [email protected] Email: [email protected]

Abstract— Econometricians deal with a tremendous amount learning models, are investigated and their field of
of data to derive the relationships between economic entities. application is also observed. In general terms, econometrics
2021 6th International Conference on Inventive Computation Technologies (ICICT) | 978-1-7281-8501-9/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICICT50816.2021.9358514

When statistical techniques are applied to the economic data to and machine learning are both techniques that employ
determine the relative economic entities with verifiable statistical and mathematical algorithms to analyze data.
observations, this quantitative analysis is termed Econometrics.
Traditional Econometric methods employ pure statistical and
Econometrics is considered to be a traditional method while
mathematical concepts to analyze economic data. Applied Machine Learning methods are considered as modern and
Econometrics deals with exploring real-world observations like extendible techniques since it evolves consistently. Several
forecasting, fluctuating market prices, economic outcomes or hybrid and novel algorithmic approaches have achieved
results, etc. In recent years, Machine Learning models are better prediction accuracies.
applied to quantitative data available in almost all domains.
Machine Learning Models perform very efficiently in the
classification process and it is used in the field of economics to
classify the economic data more accurately than traditional
econometric models. In this paper, several machine learning
methods that are specifically used for economic data are
explored. This paper further investigates the various Supervised
machine learning techniques that contribute effectively along
with metrics that are involved in the analysis procedure of
econometric models. This study provides deep insight into those
machine learning models preferred by the Econometricians and
their future implications.

Keywords— Econometrics, Economic Data, Machine Figure 1. Machine Learning Models and different types of data
Learning, Supervised, Unsupervised
In 1910, an economist named Pawel Ciompa coined the word
I. INTRODUCTION
“econometrics” and the father of econometrics is Jan
Data Science is a huge field of study that involves machine Tinbergen. Econometric uses various models in the process
learning techniques to explore various types of data like of analyzing economic data. Multiple linear regression and
Economic data, Bioinformatic data, Cheminformatic data, Ordinary Least Squares (OLS) are the two frequently used
Geo-informatics data, etc. When classification or prediction econometric techniques. The main goal of Econometricians
of data has to be performed, the first choice of all researchers is to determine such an estimator that can possess certain
would be Machine Learning algorithms. Machine Learning desirable statistical properties like consistency, efficiency,
(ML) models achieve high prediction accuracy with almost and unbiasedness. The limitation of econometrics is that
any type of quantitative data regardless of the subject and certain econometric models may end up with a result that
classify the data in both binary and multinomial way gives a relationship that is spurious among two variables [1].
efficiently. Fig. 1 shows a block diagram of Machine learning Machine learning algorithms have proved effective for
models and various types of data. When qualitative or economic data like future product designing or building,
categorical data are involved in the dataset, they are prediction of customers preferences, predict fluctuating
converted to numerical data during the pre-processing phase market trends, surveys and analyses, etc. The disadvantage of
before the model training phase. Another advantage with the econometric models is the possibility of overfitting and
ML models is that the hyperparameters can be finely tuned to forecasting errors [2]. Forecasting can be performed using
increase the prediction. Once all the basic learning models are both Econometric, and Machine Learning models.
trained, the researcher can identify the best suitable model for Econometric models compute forecasting using exponential
the dataset in particular. Economic data are enormous and smoothing models, regression models and ARIMA (Auto-
more quantitative in nature. When the dataset is balanced and Regressive Integrated Moving Average) model. Machine
limited there is less possibility for biasedness, where the basic Learning models include Regression models, Classification
ML models can be utilized. With a huge and unbalanced models, Ensemble models, and ANN (Artificial Neural
datasets basic models might not yield expected results, and Network) Models. Researchers working with economic data
hence advanced ML models like deep learning are used. In have found that ANN methods achieve higher prediction
this paper, both basic learning models, as well as advanced

978-1-7281-8501-9/21/$31.00 ©2021 IEEE 1010

Authorized licensed use limited to: Pontificia Universidad Javeriana. Downloaded on February 19,2022 at 01:19:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Sixth International Conference on Inventive Computation Technologies [ICICT 2021]
IEEE Xplore Part Number: CFP21F70-ART; ISBN: 978-1-7281-8501-9

accuracy results when compared to the traditional and NN learning algorithms. Among the three ML
econometric models [3]. techniques, RF proved to be consistently efficient. Stochastic
Gradient Descent (SGD) was used for optimizing. ReLU was
II. RELATED WORK used as an activation function and the loss function was
defined by cross-entropy [10]. Mojtaba Nabipour et al
Xianghui yuan et al obtained data for eight years from
analyzed and predicted trends in the Stock market by
China Stock Market. 60 features were taken for the study
employing nine machine learning models. Adaboost
initially and were reduced using Recursive Feature
(Adaptive Boosting), eXtreme Gradient Boosting (XG
Elimination (RFE). They selected the best features and then
Boost), Support Vector Classifier (SVC), K-Nearest
employed Machine learning Models. The Stock Market, in
Neighbour (KNN), Artificial Neural Network (ANN), Naïve
general, is dynamic and complex. Predicting the trends in
Bayes (NB), and Logistic Regression (LR) were the basic
rising and fall of prices is a huge challenge. They designed
Machine learning algorithms used. Deep learning models like
and integrated long-term Stock selection models to predict
Recurrent Neural Network (RNN) and LSTM (Long Short-
the trending Stock price. They used three machine learning
Term Memory) were also employed. Data from 4 different
techniques Support Vector Machine (SVM), Random Forest
groups were taken for the study and 10 years of data were
(RF), and Artificial Neural Network (ANN). Linear Function,
considered for the investigation. The deep learning
Polynomial, and Gaussian Kernels were used for SVM.
techniques RNN and LSTM performed well compared to the
Metrics like Accuracy and Precision were computed for
other Machine learning models with 86% accuracy [11]. Jose
Model evaluation. Cross-Validation was performed using
Ramon Saura et al compared the traditional methodology for
Time-Sliding Window Method. Random Forest was the best
financial analysis with the machine learning technique. The
performing model among the three [4]. Jose Maria et al
machine learning algorithm SVM was applied and it proved
employed unsupervised learning techniques to analyze the
better than the traditional method [12].
evolution that occurred in the Spanish Labour market.
Unsupervised Machine Learning algorithms from Spark like III. METHODOLOGY
K-Means and from Stata, the average linkage Method were
The traditional Econometric models used in Economics
applied. The optimal number of clusters was selected using
has got their own advantages and disadvantages. The main
the Internal and External Clustering validity indices method
limitation of the existing Econometric models is that it
available through K-Means from Spark. WSSSE, Davies-
sometimes results in overfitting of the samples and also
Bouldin, BD-Dunn, BD-Silhouette are some of the internal
produces forecasting errors. These models are very efficient
validity indices [5]. Xueqian Fu et al optimized distribution
with limited datasets but with Bigdata available, Machine
networks with the stochastic techniques using machine
learning models. The Unsupervised technique for clustering,
K-Means was used and PCA was employed to reduce the
dimension of the features [6]. Petr Hajek et al proposed a
backorder prediction system that increases the inventory
profit function. They transformed the dataset to a balanced
unbiased dataset using SMOTE, RUS, and Easy Ensemble.
They applied various Supervised machine learning
Classifiers like C4.5 DT (Decision Tree), SVM (Support
Vector Machine), LR (Logistic Regression), NN (Neural
Network), K-NN (K Nearest Neighbor), RF (Random
Forest), and XG BOOST. Random Forest was the best
performing model and the Decision Tree model was the next
efficient model [7]. Liang Li et al explored the machine
learning techniques that efficiently analyze investment with
intelligent procedures. Unsupervised machine learning model
K-Means algorithm was used to study the investment
efficiency, which is the most crucial factor for the
measurement of intelligent investment products [8]. Shuai
Liu et al Forecasted the Seasonal Time Series trends using
CNN (Convolutional Neural Network). They used optimizers
like Adadelta, AdaGrad, Adam, SGD, etc. Activation
functions like ReLU (Rectified Linear Unit), tanh, and linear
were applied [9].
Pablo Massaferro et al proposed a design that effectively
increased the economic return. Many people involved in the
illegal consumption of electric power that certainly has an
impact on the economy of a country. Non-technical losses
(NTL) in India, approximately was higher than $4.5 billion.
This has become a common economic development barrier in
most of the developing countries. The researchers provide an
optimized solution in terms of NTL. They applied SVM, RF, Figure. 2 The Machine Learning Approach to Economic Data

978-1-7281-8501-9/21/$31.00 ©2021 IEEE 1011

Authorized licensed use limited to: Pontificia Universidad Javeriana. Downloaded on February 19,2022 at 01:19:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Sixth International Conference on Inventive Computation Technologies [ICICT 2021]
IEEE Xplore Part Number: CFP21F70-ART; ISBN: 978-1-7281-8501-9

Learning Models would be the best option for prediction or same model can be applied to train and test a similar type of
analysis. Fig. 2 presents an approach where the prediction dataset in the future.
accuracy can be optimized using a combination of
IV. MODELS
Econometric models, Feature Engineering, and Machine
Learning Models. The methodology can be divided into six A. ECONOMETRIC MODELS
phases.
Classical and modern Econometrics are two types of
A. Data Selection Econometrics. Classical Econometrics includes single and
simultaneous equations. Modern Econometrics includes:
Relevant data is collected from the economic repositories
or economic survey reports. The standard dataset is required ➢ The Time Series Model
for the effective analyses of the data. Primary datasets and ➢ Exponential Smoothing Model
secondary datasets obtained from authorized organizations ➢ The Random Walk Model
are considered for the investigation. The primary dataset ➢ ARIMA
helps in observing new patterns. Different type of dataset ➢ Auto-Regressive Model
might have different types of best-performing machine ➢ Fitting Model
learning models when the dataset is trained using them. Moving Average, Exponential Smoothing and Holt-Winters
Logistic regression models achieve higher accuracy on smoothing are some of the time Series forecasting techniques.
certain datasets while Multi-Layered Perceptron might work [3].
well with another dataset. B. MACHINE LEARNING MODELS
B. Data Refinement I. Artificial Neural Network- Multi-Layer
Perceptron (MLP)
The primary dataset is likely to have redundant
The Multi-Layer Perceptron is an ANN that has n inputs
observations and there is also the possibility of missing data.
and m outputs. In addition to the input and output layers, there
The data can be qualitative and quantitative in nature. During
is another layer called the hidden layer which can have more
this stage, the missing data are filled and the categorical data
than one non-linear layer and are present between the input
are converted to a numerical value since statistical or machine
and the output layers. With the weighted linear summation
learning models work only with numerical data.
technique, the hidden layer transforms data from the previous
C. Application of Econometric Models layers. There can be any number of hidden layers. The output
layer generates the output by transforming the data from the
Econometric models like Multi-Linear Regression
hidden layers using the final activation function. MLP uses
(MLR), Least Square (OLS) model, ARIMA. These
backpropagation technique to compute the gradients for the
Econometric models provide the researcher with highly
gradient descent method and hence reduces the cross-entropy
collinear observations and project various errors. Irrelevant
loss function. Recurrent Neural Network (RNN) and
and redundant data can be easily identified using
Convolutional Neural Network (CNN) are also popular ANN
Econometric models and these data faults can be rectified
techniques. Deep learning techniques are the most frequently
using Feature Engineering.
used these days and it encompasses several hidden layers. In
D. Feature Engineering a feed-formed ANN architecture many activation functions
can be employed namely:
Feature Engineering is a very important phase where
• Sigmoid Function (0,1)
irrelevant features are eliminated using various feature
reduction models like Principal Component Analysis (PCA), • Rectified Linear Unit (ReLU) [0, ∞)
Recursive Feature Elimination (RFE), SelectFromModel • Hyperbolic Tangent Function (tanh) (-1,1)
(SFM), Linear Discriminant Analysis (LDA), etc. • e-function (0, ∞)
Dimensionality Reduction results in higher prediction • Softmax function (-∞, 8)
accuracy as not every feature considered for the study carries
equal importance. Deep learning techniques have proved very efficient for the
econometric data in the prediction process [14]. Some of the
E. Application of Machine Learning Models Deep learning models that are used for economic data are:
Depending upon the type of dataset-specific ML model • Long-Short Term Memory (LSTM)
can be applied. For the balanced and limited datasets, basic • Gated Recurring Unit (GRU)
ML models perform well. When the dataset is of high • Extreme Learning Machines (ELM)
dimensionality, Artificial Neural Networks have higher
prediction accuracy. More specifically, Deep Learning yields II. Regression Model – Logistic Regression (LR)
higher prediction accuracy than any other model. Logistic Regression is a classification technique that
classifies based on a logistic function which is known as the
F. Select the best ML Model sigmoid function. The performance of the Logistic Regression
The dataset is trained on all the Machine Learning models on economic data is comparatively less than other machine
and the best performing ML model is selected. The prediction learning models. This algorithm is used for both binary and
accuracy can be further enhanced by tuning the multinomial classification. This facilitates the computation of
the minimize loss and the gradient [13].
hyperparameters. When the best ML Model is selected, the

978-1-7281-8501-9/21/$31.00 ©2021 IEEE 1012

Authorized licensed use limited to: Pontificia Universidad Javeriana. Downloaded on February 19,2022 at 01:19:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Sixth International Conference on Inventive Computation Technologies [ICICT 2021]
IEEE Xplore Part Number: CFP21F70-ART; ISBN: 978-1-7281-8501-9

III. Classification Model- Decision Tree (DT) VII. Penalization Methods

DT is a supervised machine learning algorithm that is The penalization methods in machine learning are Ridge,
non-parametric. It employs a simple decision- making norm Elastic-net, and LASSO (Least Absolute Shrinkage and
using the if-then-else condition. DT handles both numerical Selection Operator). TABLE I shows the Least Square
as well as categorical data very effectively. The model fits Criterion for the estimators. Ridge is an estimator that is of
better when the depth of the tree is deep. It uses Greedy Shrinkage-type and is similar to the James-Stein Estimator.
algorithms for optimization but the main disadvantage with The aim of ridge Regression is to minimize the Collinear
DT is that it frequently overfits the data [13]. regressors. It uses L2 Penalty. The ridge regression is given
by the equation:
IV. K-Nearest Neighbor (KNN) 𝛼̂𝑟𝑖𝑑𝑔𝑒 = (𝐴′ 𝐴 + 𝜆𝐼𝑘 )−1 𝐴′ 𝐵 (2)
The training samples are separated into groups or classes and Where λ>0 is a Shrinkage parameter.
a new class is predicted based on the voting of its neighbors. LASSO is an intermediate case that uses an L1 penalty. The
The K-NN algorithm defines a distance measurement equation for LASSO is given by:
function. This function searches the nearest training samples 𝛼̂𝑙𝑎𝑠𝑠𝑜 = 𝑎𝑟𝑔 𝑚𝑖𝑛 𝐿𝑆𝐶1 (𝛼, 𝜆) (3)
𝛼
present in the feature space [14].
The penalties L1 and L2 are linearly combined in Elastic-net
[15].
V. Support Vector Machine
Non-Linear systems are classified using SVM, which is a TABLE I Estimators and their Least Square Criterion
statistical-based model. SVM’s are preferred for binary
classification while Random Forest is used for multi-class
Classification. SVM employs a hyperplane to divide the
classes. Fig. 2 shows a SVM classification [14].

V. RESULTS AND DISCUSSION

Figure 2. Support Vector Machine
There are several metrics used to measure the efficiency of
VI. Ensemble Model- Random Forest (RF) the machine learning models as shown in TABLE II.

Regression trees that are bagged can be modified. TABLE II METRICS FOR EVALUATION OF MODELS
To minimize the variance in the estimation, the trees are
further modified. The bootstrap regression trees are
decorrelated by including extra randomness. The Random
Forest follows the following steps:
1. Select a node with a minimum value 𝑋𝑚𝑖𝑛 .
2. Choose random variables 𝑛 < 𝑝.
3. For bootstrap Sample S, 𝑠 = 1,2 ⋯ 𝑆.
a. Take a random sample from S.
b. Develop a tree till 𝑋𝑚𝑖𝑛 is reached.
i. Choose n random variable among p regressors.
ii. Among n variables, select the best node and
variable for the regression tree.
iii. Beginning from the node split the sample.
c. In the bootstrap tree, let us consider 𝑛̂𝑠 (𝑥) to be a
regression tree. 𝐼𝑗 be the sample mean on each branch. The
variance is hence reduced by the equation
1
𝑛̂𝑟𝑓 (𝑥) = ∑𝑠𝑆=1 𝑛̂𝑠 (𝑥) (1)
𝑆
The number of variables was reduced from p to n at each
step [15].
.

978-1-7281-8501-9/21/$31.00 ©2021 IEEE 1013

Authorized licensed use limited to: Pontificia Universidad Javeriana. Downloaded on February 19,2022 at 01:19:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Sixth International Conference on Inventive Computation Technologies [ICICT 2021]
IEEE Xplore Part Number: CFP21F70-ART; ISBN: 978-1-7281-8501-9

Figure 3 Comparison of Econometric and Machine Learning Models

Rounaq Basu et al explored econometric and machine There are other factors like over-fitting, under-fitting, and
learning models and compared their relative efficiency. The error metrics.
data was drawn from the Household Interview Travel Survey TABLE III Error metrics for model evaluation
(HITS) conducted in Singapore in 2012. Fig 3. Shows the
training and testing accuracy of Econometric and ML models.
Multinomial Logit (MNL) and Ordinal Logit (OL) are two
Econometric models that were considered for the study.
Ordinal Logit Classification (OLC)-Neural Network was
compared with OL and other six ML models were compared
with MNL [17]. Outliers, nature, and scale of the data are
some of the factors that affect the error metrics. In the process
of prediction, error metrics can influence ranking. Error
metrics show the difference between the actual and the
predicted value and hence the model can be evaluated for its
efficiency.

TABLE III shows various error metrics used for model

evaluation where 𝑏𝑖 is the actual value, 𝑏̂𝑖 is the estimated or
predicted value of 𝑏𝑖 . and k is the number of samples or
observations [3].

Figure 4. Model Evaluation Metrics for Econometric and ML

Models Figure 5. The Execution time of Econometric and ML Models

Fig. 4 shows evaluation metrics like precision, recall, and F- Fig. 5 shows the execution time of both the econometric
Measure. The evaluation metric plays a crucial part in models and machine learning models. MNL has the longest
determining the accuracy of the machine learning models. execution time. Clearly, Neural Network (NN) proves to be an
Accuracy alone cannot determine a model’s performance. efficient machine learning model and is also better than
econometric models.

978-1-7281-8501-9/21/$31.00 ©2021 IEEE 1014

Authorized licensed use limited to: Pontificia Universidad Javeriana. Downloaded on February 19,2022 at 01:19:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Sixth International Conference on Inventive Computation Technologies [ICICT 2021]
IEEE Xplore Part Number: CFP21F70-ART; ISBN: 978-1-7281-8501-9

TABLE IV Application of ML Models

Author Subject Data-Set ML Models Other Metrics Best
Model
Gabriel et al Energy Six decades of data Traditional econometric MAPE, RMSE, Diebold- ANN-
[17] Forecasting that includes prices of model, ANN (Multi- Mariano (M-DM) MLP
Oil, Coal, Natural gas Layered Perceptron), RF,
and was obtained from No-Change method.
the International ARIMA
Monetary Fund (IMF)
Christian Computing Data was drawn from ANN (Artificial Neural MSE (Mean Square ANN-
Behm et al Annual the Data Portal and Network)- Multi-Layered Error), 𝑅2 , MAPE (Mean MLP
[18] electricity Power Statistics and Perceptron (MLP), Mid- Absolute Percentage
load profiles the Transparency Term Adequacy Forecast Error), RMSE (Root Mean
for Platform for (MAF) Method. Square Error)
European Electricity loads.
countries
Xuerong Li Forecasting Data was observed RF, Support Vector MAE, RMSE, Accuracy, ANN-
et al [19] Crude Oil from daily West Texas Machine (SVM), Linear Precision, Recall, F- Deep
Price with Intermediate (WTI) Regression, ANN- Deep Measure. learning-
text-based Learning Technique- RFE (Recursive Feature CNN
Mining and Convolutional Neural Elimination), LDA (Latent
Machine Network (CNN) Dirichlet Allocation) for
Learning feature grouping.
Modhurima Predicting Data was retrieved RF, eXtreme Gradient MSE, Normalized Root XGB
Dey Amin et access to from the American Boost (XGB), LASSO Mean Square Error
al [20] food Community Survey (Least Absolute (NRMSE), Cohen’s Kappa
retailers 2010. Shrinkage and Selection
using Operator).
Machine
Learning
models
Jayanth Stock Price Traded Banking stock Deep Learning Models- RMSE, Directional GRU for
Balaji et al Forecasting details are obtained Long-Short Term Accuracy (DA), Median shorter
[21] using Deep from the S&P BSE- Memory (LSTM), Gated Absolute Percentage Error forecast
learning BANKEX. Recurring Unit (GRU), (MdAPE) horizon.
models Convolutional Neural ELM for
Networks (CNN), longer
Extreme Learning forecast
Machines (ELM) horizon.
Alev Dilek Compare Data was drawn from Artificial Neural Network Selection Criteria: ANN
Aydin et al the Electronic Data (ANN), Vector Auto Akaike Information
[22] prediction Distribution System of Regressive (VAR) Criterion (AIC),
of ANN and the Central Bank of Hannan-Quinn Criterion
VAR the Republic of (HQC), Schwarz
Turkey (CBRT) Information Criterion
(SIC), Final Prediction
Error (FPE)
Junhao Prediction Data was retrieved Deep Regression, RMSE, MAE, MAPE CNN
Zhou et al of Metal from Macrotrends Support Vector
[23] Price using Regression, ARIMA,
Machine RNN (Recurrent Neural
Learning Network), CNN, LSTM
Models (Long Short-Term
Memory)

978-1-7281-8501-9/21/$31.00 ©2021 IEEE 1015

Authorized licensed use limited to: Pontificia Universidad Javeriana. Downloaded on February 19,2022 at 01:19:05 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Sixth International Conference on Inventive Computation Technologies [ICICT 2021]
IEEE Xplore Part Number: CFP21F70-ART; ISBN: 978-1-7281-8501-9

Hybrid models that comprise of multiple Machine learning Analysis," in IEEE Access, vol. 8, pp. 150199-150212, 2020, doi:
models combined with statistical techniques are widely used 10.1109/ACCESS.2020.3015966.
in cloud computing environments [24]. Econometric models [12] J. R. Saura, B. R. Herráez and A. Reyes-Menendez, "Comparing a
Traditional Approach for Financial Brand Communication Analysis
and advanced machine learning algorithms efficiently With a Big Data Analytics Technique," in IEEE Access, vol. 7, pp.
contribute to Big data analytics [25]. 37100-37108, 2019, doi: 10.1109/ACCESS.2019.2905301.
[13] Basu, Rounaq & Ferreira, Joseph. (2020). Understanding household
vehicle ownership in Singapore through a comparison of econometric
VI. CONCLUSION and machine learning models. Transportation Research Procedia. 48.
10.1016/j.trpro.2020.08.207.
Econometric models are used for forecasting and there [14] Carlos A. Peña-Solórzano, David W. Albrecht, Richard B. Bassed,
Michael D. Burke, Matthew R. Dimmock, Findings from machine
are several methods that are applied for the Economic data. learning in clinical medical imaging applications – Lessons for
Econometric models are not efficient enough to identify translation to the forensic setting, Forensic Science International,
complex relationships. Machine learning overcomes the Volume-316,2020,110538,ISSN0379-0738,
https://doi.org/10.1016/j.forsciint.2020.110538.
limitations of Econometric models in terms of prediction.
[15] https://perhuaman.files.wordpress.com/2014/06/econometrics-bruce-
Machine learning models are applied further when the data is hansen-2014.pdf
huge and the structure is unknown. Among several machine [16] Basu, Rounaq & Ferreira, Joseph. (2020). Understanding household
learning models Random Forest is one of the Ensemble vehicle ownership in Singapore through a comparison of econometric
and machine learning models. Transportation Research Procedia. 48.
methods that predicts with a high accuracy, when the data is 10.1016/j.trpro.2020.08.207.
limited. Deep learning techniques proved more efficient with [17] Gabriel Paes Herrera, Michel Constantino, Benjamin Miranda Tabak,
huge economic data and by hyper-tuning the parameters Hemerson Pistori, Jen-Je Su, Athula Naranpanawa, Data on forecasting
energy prices using machine learning, Data in Brief, Volume 25, 2019,
higher accuracy was achieved. Incorporating hybrid machine 104122, ISSN 2352-3409. https://doi.org/10.1016/j.dib.2019.104122.
learning models, Stochastic time series methods and [18] Christian Behm, Lars Nolting, Aaron Praktiknjo, How to model
designing novel deep learning models would result in higher European electricity load profiles using artificial neural networks,
accuracy with economic data in future. Applied Energy,Volume 277, 2020, 115564,ISSN 0306-2619,
https://doi.org/10.1016/j.apenergy.2020.115564.
[19] Xuerong Li, Wei Shang, Shouyang Wang,Text-based crude oil price
forecasting: A deep learning approach, International Journal of
REFERENCES Forecasting, Volume 35, Issue 4, 2019, Pages 1548-1560, ISSN 0169-
[1] https://en.wikipedia.org/wiki/Econometrics 2070, https://doi.org/10.1016/j.ijforecast.2018.07.006.
[2] https://addepto.com/machine-learning-in-economics-how-is-it-used [20] Modhurima Dey Amin, Syed Badruddoza, Jill J. McCluskey,
Predicting access to healthful food retailers with machine learning,
[3] https://www.diva-portal.org/smash/get/diva2:1312334
Food Policy,2020,101985,ISSN 0306-9192,
[4] X. Yuan, J. Yuan, T. Jiang and Q. U. Ain, "Integrated Long-Term Stock
https://doi.org/10.1016/j.foodpol.2020.101985.
Selection Models Based on Feature Selection and Machine Learning
[21] A. Jayanth Balaji, D.S. Harish Ram, Binoy B. Nair,Applicability of
Algorithms for China Stock Market," in IEEE Access, vol. 8, pp.
Deep Learning Models for Stock Price Forecasting An Empirical Study
22672-22685, 2020, doi: 10.1109/ACCESS.2020.2969293.
on BANKEX Datab, Procedia Computer Science, Volume 143, 2018,
[5] J. M. Luna-Romera, F. Núñez-Hernández, M. Martínez-Ballesteros, J. Pages947-953,ISSN1877-0509,
C. Riquelme and C. Usabiaga Ibáñez, "Analysis of the Evolution of the https://doi.org/10.1016/j.procs.2018.10.340.
Spanish Labour Market Through Unsupervised Learning," in IEEE
[22] Alev Dilek Aydin, Seyma Caliskan Cavdar,Comparison of Prediction
Access, vol. 7, pp. 121695-121708, 2019, doi:
Performances of Artificial Neural Network (ANN) and Vector
10.1109/ACCESS.2019.2935386.
Autoregressive (VAR) Models by Using the Macroeconomic Variables
[6] X. Fu, Q. Guo and H. Sun, "Statistical Machine Learning Model for of Gold Prices, Borsa Istanbul (BIST) 100 Index and US Dollar-
Stochastic Optimal Planning of Distribution Networks Considering a Turkish Lira (USD/TRY) Exchange Rates,Procedia Economics and
Dynamic Correlation and Dimension Reduction," in IEEE Finance,Volume 30,2015,Pages 3-14, ISSN 2212-5671,
Transactions on Smart Grid, vol. 11, no. 4, pp. 2904-2917, July 2020, https://doi.org/10.1016/S2212-5671(15)01249-6.
doi: 10.1109/TSG.2020.2974021.
[23] J. Zhou et al., "Precious Metal Price Prediction Based on Deep
[7] P. Hajek and M. Z. Abedin, "A Profit Function-Maximizing Inventory Regularization Self-Attention Regression," in IEEE Access, vol. 8, pp.
Backorder Prediction System Using Big Data Analytics," in IEEE 2178-2187, 2020, doi: 10.1109/ACCESS.2019.2962202.
Access, vol. 8, pp. 58982-58994, 2020, doi:
[24] S.Nikkath Bushra, Dr.A.Chandra Sekar, “An Incremental Clustering
10.1109/ACCESS.2020.2983118.
Technique For Privacy Preservation Over Incremental Cloud Data”,
[8] L. Li, J. Wang and X. Li, "Efficiency Analysis of Machine Learning International Review On Computers And Softwares(I.RE.CO.S),
Intelligent Investment Based on K-Means Algorithm," in IEEE Access, Vol.8,N.11,ISSN 1828-6003, November 2013.
vol. 8, pp. 147463-147470, 2020, doi:
[25] S.Nikkath Bushra, Dr.A.Chandra Sekar,” Descend Traversal
10.1109/ACCESS.2020.3011366.
Prioritization and Ascend Traversal Abstraction Based Data
[9] S. Liu, H. Ji and M. C. Wang, "Nonpooling Convolutional Neural Anonymization Over Big Data”, International Journal Of Applied
Network Forecasting for Seasonal Time Series With Trends," in IEEE Engineering Research Issn 0973-4562 Volume 10, Number 2 (2015)
Transactions on Neural Networks and Learning Systems, vol. 31, no. Pp. 2979-2994.
8, pp. 2879-2888, Aug. 2020, doi: 10.1109/TNNLS.2019.2934110.
[10] P. Massaferro, J. M. D. Martino and A. Fernández, "Fraud Detection in
Electric Power Distribution: An Approach That Maximizes the
Economic Return," in IEEE Transactions on Power Systems, vol. 35,
no. 1, pp. 703-710, Jan. 2020, doi: 10.1109/TPWRS.2019.2928276.
[11] M. Nabipour, P. Nayyeri, H. Jabani, S. S. and A. Mosavi, "Predicting
Stock Market Trends Using Machine Learning and Deep Learning
Algorithms Via Continuous and Binary Data; a Comparative

978-1-7281-8501-9/21/$31.00 ©2021 IEEE 1016

Authorized licensed use limited to: Pontificia Universidad Javeriana. Downloaded on February 19,2022 at 01:19:05 UTC from IEEE Xplore. Restrictions apply.