Received 10 July 2022, accepted 29 July 2022, date of publication 1 August 2022, date of current version 8 August 2022.

Digital Object Identifier 10.1109/ACCESS.2022.3195942

Decision Fusion for Stock Market Prediction: A

Systematic Review
CHENG ZHANG , (Member, IEEE), NILAM N. A. SJARIF,
AND ROSLINA B. IBRAHIM , (Member, IEEE)
Advanced Informatics Department, Razak Faculty of Technology and Informatics, Universiti Teknologi Malaysia, Kuala Lumpur 54100, Malaysia
Corresponding author: Cheng Zhang ([email protected])

ABSTRACT Stock market prediction based on machine or deep learning is an essential topic in the financial
community. Typically, models with different structures or initializations provide different forecasts of the
same response variable. In such cases, better prediction is often achieved by combining forecasts from
multiple models rather than using a single model in isolation. This combination of forecasts from the base
learners is known as decision fusion. Furthermore, although decision fusion is typical and essential for
making the best possible use of multiple forecasts, few studies have systematically summarized the studies
that apply this technique. Therefore, there is an urgent need for a literature review reflecting the application of
decision fusion in this field. To this end, this study systematically reviewed research related to decision fusion
for stock market prediction, focusing on the characteristics of base learners and decision fusion methods.
Specifically, the research trend on this topic, which has shifted over the past two decades, is discussed. This
review also presents future directions in applying decision fusion to stock market prediction, such as the
fusion of forecasts with different data types, using new algorithms as base learners, and integrating sentiment
analysis with decision fusion techniques.

INDEX TERMS Base learner, decision fusion, ensemble, machine learning, review, stock market prediction.

I. INTRODUCTION tion. In such cases, better prediction is often achieved by com-

Stock market prediction is essential to the financial commu- bining multiple forecasts rather than by using a single model
nity and helps develop effective security trading strategies in isolation [4]. This ‘‘combining the wisdom of crowds’’
[1]. Machine learning-based stock market forecasting usually approach is known as ensemble learning, which involves
involves applying a machine learning algorithm to learn a pat- training several base learners separately and combining their
tern from historical data and then predict the stock market’s forecasts [5]. Ensemble learning applies decision fusion, or in
future. With rapidly increasing computing power over the past other words, merges the ‘‘decisions’’ of several base learners
20 years and the massive amount of data available from the into a single ‘‘decision’’ about the response variable [6].
Internet, machine learning algorithms have begun to show Contextually, the term ‘‘fusion’’ means integrating data or
advantages in forecasting [2]. Li and Bastos [3] reviewed knowledge from multiple sources [7] and can be classified
the latest research on stock market prediction using deep into three types: data fusion, feature fusion, and decision
learning techniques from 2017 to 2020. They found that deep fusion [8]. Therefore, this combination of multiple forecasts
learning models also play a significant role in stock market provided by base learners is also referred to as decision fusion
forecasting. [9], [10].
Typically, given a response variable, different forecasts can Fusion for stock market prediction involves several areas,
be generated if the models used for prediction have different such as artificial intelligence, data fusion techniques, and
structures or the same structure but with random initializa- finance, and there is no fixed approach on how to use fusion
techniques. Owing to the diversity of stock market forecasts,
The associate editor coordinating the review of this manuscript and the choice of decision fusion method usually varies depend-
approving it for publication was Mostafa M. Fouda . ing on the response variable and individual preferences.

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
81364 VOLUME 10, 2022
C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

Nevertheless, the fundamental principle is that the final pre- trast, other reviews only include studies on forecasting stock
diction should be generated based on the perceived situational market trends [14], [16]. On the other hand, two reviews have
knowledge [11]. If multiple forecasts are fully exploited, then focused on broader areas, such as stock market price/trend
more valuable predictions can be obtained at the decision and risk/return forecasting [21], and even predictions in areas
level [12]. from the stock market and e-commerce to corporate banking
However, although decision fusion is typical and essential and cryptocurrency [17].
for making the best possible use of multiple forecasts [13], Most reviews mainly focused on the application of
few studies have systematically summarized the studies that machine learning and deep learning techniques [14]–[17],
apply this technique. In this context, there is an urgent need [19]. This phenomenon can be attributed to the rapid increase
for a literature review that presents research trends on this in computing power in recent years, which has significantly
topic and provides clues to researchers who wish to apply affected the stock market prediction. Researchers have found
decision fusion in their studies. Consequently, this paper that commonly used machine learning algorithms for effec-
presents a systematic review of research related to decision tive prediction include artificial neural networks (ANNs),
fusion for stock market prediction, considering relevant stud- fuzzy-based techniques [14], and support vector machines
ies published in two scientific databases (Scopus and Web (SVMs) [18]. Nosratabadi et al. [17] stated that deep learning
of Science). For each proposed model from the included algorithms, particularly long short-term memory (LSTM),
studies, the focus was on the fundamental aspects of the deci- convolutional neural networks (CNNs), and deep neural net-
sion fusion process, including the base learners and decision works (DNNs), are the most applied techniques for analyzing
fusion methods. The main content is categorized based on financial time-series data. However, Pandurang and Kumar
whether the forecasting task is classification or regression. [15] and Bustos and Pomares-Quimbaya [16] reported that
The remainder of this paper is organized as follows. ‘‘single-handed’’ models, such as SVM and ANN, are not
Section II analyzes the relevant review work and describes as efficient as other hybrid ensembles. This finding indicates
the review methodology. Section III provides an overview of that decision fusion is a promising approach for improving
the studies included in this review. Section IV summarizes the prediction.
characteristics of the base learners in each forecasting model Unfortunately, few reviews have considered decision
from the included studies. Section V extracts the fusion meth- fusion techniques. Instead, more attention was given to the
ods used at the decision level. Section VI provides an overall raw datasets used in the included studies. Nti et al. [18] cal-
analysis of the included studies, categorizing the forecasting culated the number of data sources and found that nearly 90%
models based on their structures. Finally, this study provides of studies favored a single data source. Moreover, Nalabala
some concluding remarks and highlights several directions and Nirupamabhat [20] focused on data-mining techniques
for future research. and sentiment analysis. They argued that analyzing and sum-
marizing data with opinions is helpful for better predictions.
II. RELATED WORK AND RESEARCH METHODOLOGY Unlike the authors above, Thakkar and Chaudhari [21] focus
The first step in this study is to evaluate review studies on data and feature fusion techniques for stock market predic-
related to decision fusion in the field of stock market pre- tions. They argued that the unification of widespread classes
diction. Two leading databases of scientific papers, Web of given by various classification categories could be a contex-
Science and Scopus, were selected to search for related work. tual representation for future work.
Furthermore, only articles with ‘‘review’’ or ‘‘survey’’ as Although most review articles do not focus primarily
keywords or partial abstracts were considered for discarding on fusion techniques, they still refer to ensemble models
non-survey or non-review papers. Another filtering criterion that employ decision fusion [14]–[19]. Other studies have
was the presence of the words (‘‘fusion’’ or ‘‘integration’’ or also used ‘‘fusion models’’ to describe decision fusion [20],
‘‘aggregation’’ or ‘‘combination’’) and (‘‘stock market’’ or [21]. Therefore, it can be concluded from the comparative
‘‘financial market’’) and (‘‘forecast’’ or ‘‘prediction’’) in the analysis of these reviews that researchers have realized the
abstract or keywords. Eight articles were identified after the high-quality performance of ensemble or fusion models for
screening. stock market prediction.
Table 1 presents a comparative analysis of relevant review Once similar studies were evaluated, the search equation
studies under stock market prediction and fusion criteria. was defined as searching for studies using decision fusion
The columns indicate several aspects of each review paper, for stock market prediction. In this review, Web of Science
including the number of references and citations, primary and Scopus databases were searched. The search period was
concerns, and content related to fusion. These articles are from the start of the database’s record availability to October
helpful to the scientific community because the number of 27, 2021. As base learners and fusion methods are two fun-
references for these studies exceeds 400; therefore, these damental aspects of the decision fusion process, this review
studies present, to some extent, the latest research trends in answers the following research questions: (1) What are the
stock market prediction. characteristics of the base learners? This study examines the
As shown in Table 1, four reviews surveyed studies on fore- base learners required to constitute the overall forecasting
casting stock market price or trend [15], [18]–[20]. In con- model and classifies them according to their heterogeneity

VOLUME 10, 2022 81365

 C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

TABLE 1. A comparative analysis of related review work under various criteria.

TABLE 2. Major journal rankings.

FIGURE 1. Flow diagram of the selection process.

and forecasting tasks. (2) Which fusion methods are applied web tool designed to help researchers working on systematic
at the decision level? This study extracts the fusion methods reviews, and the tool used for full-text screening and synthesis
used at the decision level for different forecasting tasks. is NVivo. A total of 642 articles met the search criteria,
The following inclusion criteria were considered when and 195 duplicate articles were excluded. After screening
selecting these studies: IC1: the included study predicted the by title and abstract, another 230 articles were excluded
stock market using machine learning or deep learning, and because decision fusion was not applied. Next, 38 articles
IC2: Decision fusion was used in the included study. were excluded because their full text could not be retrieved.
The search query included a set of keywords: TITLE- After a thorough reading of the remaining articles, 104 studies
ABS-KEY ((((’’ decision’’ OR ‘‘decision level’’ OR were excluded for the following reasons:1) unrelated to stock
‘‘decision-level’’) AND (‘‘fusion’’ OR ‘‘integration’’)) OR market prediction (n=40); 2) lack of details about decision
‘‘group decision-making’’ OR ‘‘group decision making’’ OR fusion or no decision fusion (n=34); 3) lack of samples,
‘‘ensemble’’ OR ‘‘combination’’) AND (‘‘stock market’’ results, or fusion methods (n=10); 4) comparative studies
OR ‘‘financial market’’ OR ‘‘stock exchange’’ OR ‘‘equity (n=9); 5) similar content (n=7); and 6) not based on machine
market’’ OR ‘‘share market’’ OR ‘‘financial price model’’ learning (n=4). After the selection process, seventy-five arti-
OR ‘‘financial volatility’’) AND (‘‘forecast∗’’ OR ‘‘stock cles were included in this review.
return’’ OR ‘‘predict∗’’ OR ‘‘forecast performance’’ OR
‘‘price model’’ OR ‘‘algorithm’’ OR ‘‘computational intel- III. OVERVIEW
ligence’’ OR ‘‘machine learning’’ OR ‘‘time series analysis’’ This section provides an overview of the studies included
OR ‘‘big data’’)). in this review. Most of the included papers were published
The selection process is illustrated in Fig. 1. The tool by indexed journals; the major journal rankings are listed
used for the title and abstract screening is Rayyan, a free in Table 2. Expert Systems with Applications, Applied Soft

81366 VOLUME 10, 2022

C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

TABLE 3. Homogeneous base learners for classification.

FIGURE 2. Number of publications per year.

Computing, Applied Intelligence and Fluctuation, and Noise

Letters are the most relevant journals. The other reviewed TABLE 4. Homogeneous base learners for regression.

papers have been presented at international conferences.

Fig. 2 provides an overview of the number of papers pub-
lished in different years. The number of articles that met
the selection criteria has increased annually, indicating the
popularity of machine learning-based stock market prediction
and decision fusion techniques. Fig. 3 shows the statistics of
the datasets and the input attributes. The US stock market
was the most investigated, well ahead of the stock markets
in other countries. Next, datasets of various lengths were
used, with periods between 5 and 10 years being the most
popular. In addition, more than half of the studies focused
on regression tasks, particularly for forecasting stock index
prices, followed by binary and multi-class classification. For
input attributes, a large proportion of the studies chose only
historical trading data, followed by the combination of histor-
ical data and technical indicators. More details of each study
are provided in Table 9 in the Appendix.

IV. CHARACTERISTICS OF BASE LEARNERS

Individual learners strategically grouped in an ensemble or
fusion model are called the base learners [97]. The perfor-
mance of an ensemble or fusion model depends heavily on the
characteristics of the base learners, such as which algorithm
each base learner employs or whether the base learners in a
forecasting model are homogeneous or heterogeneous. The networks (PNNs) [34], extreme learning machines (ELMs)
concept of heterogeneity of base learners is shown in Fig. 4. [48], and deep belief networks (DBNs) [84].
In this section, the included studies are divided into four Table 4 summarizes the studies that applied homogeneous
groups based on base learners’ heterogeneity and the category base learners for regression tasks. Again, ANN was the most
of forecasting tasks. The forecasts generated by the base popular, followed by LSTM, decision trees, and support vec-
learners are also discussed because they affect the selection tor regression (SVR). Tables 3 and 4 show that ANN, decision
of the fusion methods. trees, SVM, and LSTM are the most popular algorithms used
by homogeneous base learners regardless of the forecast-
A. HOMOGENEOUS BASE LEARNERS ing task. Moreover, the base learner can be a hybrid algo-
As shown in Fig. 4, the base learners in an ensemble or rithm such as an exponential autoregressive conditional het-
fusion model are described as homogeneous if they use the eroskedasticity backpropagation neural network (EGARCH-
same algorithm. Table 3 summarizes the studies that applied BPNN) [54] or CNN-LSTM [76].
homogeneous base learners for classification tasks. We found
that ANN was the most commonly used algorithm, followed B. HETEROGENEOUS BASE LEARNERS
by decision trees, SVM, and LSTM. In addition to these Heterogeneous base learners refer to situations in which one
algorithms, base learners can also be probabilistic neural ensemble or fusion model consists of a set of base learners,

VOLUME 10, 2022 81367

 C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

FIGURE 3. Statistics of datasets and input attributes.

FIGURE 4. Heterogeneity of base learners: a) Homogeneous base

learners; b) Heterogeneous base learners.

each of which employs a different algorithm. The purpose of

using different algorithms is to ensure ensemble diversity. FIGURE 5. Taxonomy of forecasts of base learners.
Table 5 summarizes studies that applied heterogeneous
base learners for classification tasks. In addition to using
traditional machine-learning algorithms, base learners can
use non-machine learning methods, such as user knowledge ing algorithms, non-machine learning methods, such as the
[56] and speech and text encoders [77]. It is rare for the two Delphi method [51], have also been used.
forecasting models to employ the same set of base learners.
Table 6 summarizes the studies that employed heteroge- C. FORECASTS OF BASE LEARNERS
neous base learners for regression tasks. Of the 75 articles Fig. 5 presents the taxonomy of forecasts generated by the
reviewed, only one study did not provide details about base base learners in the included studies. For binary classifica-
learners, but the authors mentioned cases in which the base tion, the forecasts of the base learners were mainly stock
learners were heterogeneous [94]. Apart from machine learn- movement trends (e.g., up and down) and stock movement

81368 VOLUME 10, 2022

C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

TABLE 5. Heterogeneous base learners for classification. TABLE 7. Studies related to classification tasks.

TABLE 8. Studies related to regression tasks.

TABLE 6. Heterogeneous base learners for regression.

Tables 7 and 8 summarize the studies by the forecasts of

the base learners. Several studies have focused on binary
classification, whereas only a few have considered multi-
class classifications. However, most regression tasks refer
to predicting stock prices or intrinsic mode function (IMFs)
values. In addition, the final forecast can be the fusion result
of different types of predictions, such as the fusion of stock
movement trends and external factors [56], the fusion of stock
prices and external factors [51], the fusion of stock movement
probability and external factors [78], and the fusion of stock
volatility and external factors [50].

V. DECISION FUSION METHODS

probability. In contrast, multi-class classification refers to Admittedly, a better prediction can be obtained by fusing
forecasting stock turning indicators, the confidence level of multiple forecasts of the base learners. However, the choice
stock movement, stock risk level, or stock movement (e.g., of the fusion method is also critical to the performance of
up, neutral, and down). Similarly, the regression forecasts are the entire model. In addition, the selection of fusion methods
diverse and include future stock prices, stock returns, and often depends on whether the forecasting task is classification
volatility. or regression. Figs. 6 shows the proposed taxonomies of the

VOLUME 10, 2022 81369

 C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

FIGURE 6. Taxonomy of decision fusion methods: a) methods for classification; b) methods for regression.

FIGURE 7. Fusion methods for classification. FIGURE 8. Fusion methods for regression.

fusion methods that the included studies used at the decision B. FUSION METHODS FOR REGRESSION
level for classification and regression. Fig. 8 shows the main decision fusion methods employed
for regression. Interestingly, the simple average, or arithmetic
A. FUSION METHODS FOR CLASSIFICATION average of all forecasts generated from the base learners, was
Fig. 7. shows the main methods used to fuse the classifi- used the most. This result was consistent with the findings of
cation results. Voting, especially majority voting, and tree- Genre et al. [100]. Here, a simple average can be perceived as
based methods are the two most commonly used methods. a specific case of weighted arithmetic mean with equal weight
Other voting methods include accuracy-based voting [29], for each component. Models that employ summation at the
consistent voting [73], plurality voting [84], and weighted decision level refer to situations in which forecasts generated
voting [39]. Here, the voting method makes a collective deci- by all base learners must be summed. Therefore, each forecast
sion from several base learners[98]. of the base learner is a part of the final prediction. This type
The second most popular method for fusing classification of model primarily refers to decomposition-based ensembles.
results is the tree-based method, which involves first feeding The forecasts of the base learners are the future IMFs values,
forecasts of all base learners into a tree-based algorithm, and the final prediction is the summation of all the predicted
then mapping each prediction to a neighborhood in the set IMFs values [101]. More sophisticated methods, such as
of dependent variables, and then returning the mean neigh- ANN and stacking (meta-learning), have also been employed
borhood [99]. The most commonly used tree-based methods to fuse regression forecasts [50]–[55], [78].
include gradient boosting [40], [75], [96] and random forest
[28], [40], [75]. It is worth noting that Barak et al. [27] used VI. OVERALL ANALYSIS
five tree-based methods for decision fusion: the BF tree, deci- After analysis of the characteristics of base learners and
sion table, decision tree, decision tree naïve Bayes (DTNB), the fusion methods, it is clear that the structures of the
and the LAD tree, with the decision table performing models proposed in the included studies follow several pat-
the best. terns. Based on their design, these forecasting models can be

81370 VOLUME 10, 2022

C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

FIGURE 9. Data pipelines of decision fusion models.

roughly categorized into four types:1) traditional ensemble, popular only after 2015. The difference between traditional
2) decomposition-based ensemble, 3) fusion models integrat- and decomposition-based ensembles is that the former’s base
ing auxiliary forecasting, and 4) two-stage ensemble. learners produce complete forecasts of the response variable.
Fig. 9 shows the data pipelines of the four types of decision The latter base learners only predict parts of the response vari-
fusion models. Most forecasting models fall under the cate- able, with the final forecast being the sum of these forecasts.
gory of traditional ensembles. This type of model typically The third type of model is the fusion model combining
has more than two base learners, each producing one forecast central and auxiliary forecasts. The latter is often a forecast
for the response variable. The traditional ensemble has been of the effect of external factors generated by either a machine
used for stock market prediction since 2000, and the number learning model [50] or non-machine learning methods, such
of relevant studies has increased annually. The second type as the Delphi method [51] or expert knowledge [56]. The
of model was decomposition-based ensembles. This type original forecast is refined at the decision fusion stage by
of model was exclusive to the regression task and became integrating the effects of the external factors. This type of

VOLUME 10, 2022 81371

 C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

TABLE 9. Bibliography of stock market prediction project. Not available (NA), Open (O), High (H), low (L), Close (C), Volume (V).

81372 VOLUME 10, 2022

C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

TABLE 9. (Continued.) Bibliography of stock market prediction project. Not available (NA), Open (O), High (H), low (L), Close (C), Volume (V).

VOLUME 10, 2022 81373

 C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

TABLE 9. (Continued.) Bibliography of stock market prediction project. Not available (NA), Open (O), High (H), low (L), Close (C), Volume (V).

fusion model can be seen as an early version of the model stage ensemble [90]. The distribution of the various decision
using decision fusion for stock market prediction because it fusion models according to the publication year is shown in
mainly appeared in the 1990s. Finally, a two-stage ensemble Fig. 10.
was recently reported in which the forecast from the first- Furthermore, based on the analysis in Section IV, for each
stage ensemble was used as the input feature for the second- forecasting model, the forecasts of the base learners always

81374 VOLUME 10, 2022

C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

TABLE 10. List of abbreviations. TABLE 10. (Continued.) List of abbreviations.

VOLUME 10, 2022 81375

 C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

TABLE 10. (Continued.) List of abbreviations. Decision fusion techniques for stock market predictions
have been developing rapidly. Researchers have shifted their
attention from the fusion of machine learning forecasts with
non-machine learning forecasts in the 1990s to traditional
ensembles since 2000 and recently to decomposition-based
ensembles.
This study had several limitations. First, relevant articles
not included in the selected databases may have been omitted.
Second, although the definition of keywords used for the liter-
ature search covers many studies, we may have missed some
studies that used less common linguistic terms. However,
to our knowledge, this review covers most forecasting models
that have applied decision fusion. In addition, the findings
presented in this review would also be insightful for other
financial prediction problems such as cryptocurrency price
or exchange rate prediction. Most importantly, these findings
will help newcomers keep on the right track when they want
to build a model that intends to produce a better prediction by
combining multiple forecasts.
Finally, for each forecasting model, the forecasts generated
by the base learners always had the same data type. Con-
sidering the diversification of forecasts in the stock market
FIGURE 10. Distribution of decision fusion models.
and the necessity of decision fusion, the unification of broad
classes or even the fusion of forecasts with different data
types is a new challenge for stock market prediction in the
had the same data type. Even when these forecasts target future. In addition, developing an ensemble with base learners
different response variables, their data type remains the same, using other algorithms, such as the jump-diffusion model,
implying that the fusion methods that appeared in this review generative adversarial network (GAN), graph neural network
can only deal with forecasts with the same data type. (GNN), and capsule network, can be a future direction for
researchers. Furthermore, only a few studies on sentiment-
VII. CONCLUSION aware prediction have applied decision fusion. The potential
This study presents a systematic review of the literature of integrating sentiment analysis and decision fusion tech-
related to decision fusion for stock market prediction. It ana- niques to improve stock market predictions can be further
lyzes two essential aspects of the decision fusion process: the exploited in future studies.
base learners and fusion methods. To this end, 75 studies were
selected and examined in detail. APPENDIX
A forecasting model that applies decision fusion can con- See Table 9 and Table 10.
sist of only homogeneous base learners, such as ANN, deci-
sion trees, SVM, and LSTM, or heterogeneous base learners REFERENCES
employing various algorithms. Two studies that used the same
[1] X. Zhou, Z. Pan, G. Hu, S. Tang, and C. Zhao, ‘‘Stock market prediction
set of heterogeneous base learners were rare. For binary on high-frequency data using generative adversarial nets,’’ Math. Prob-
classification, the forecasts of the base learners included stock lems Eng., vol. 2018, pp. 1–11, Apr. 2018, doi: 10.1155/2018/4907423.
movement trends or stock movement probabilities. In con- [2] Z. Chen, C. Li, and W. Sun, ‘‘Bitcoin price prediction using
machine learning: An approach to sample dimension engineering,’’
trast, for multi-class classification, the forecasts of base learn- J. Comput. Appl. Math., vol. 365, Feb. 2020, Art. no. 112395, doi:
ers include stock turning indicators, the confidence level of 10.1016/j.cam.2019.112395.
stock movement, stock risk level, or stock movement (e.g., [3] A. W. Li and G. S. Bastos, ‘‘Stock market forecasting using deep learn-
ing and technical analysis: A systematic review,’’ IEEE Access, vol. 8,
up, neutral, and down). Regression forecasts of base learners pp. 185232–185242, 2020, doi: 10.1109/access.2020.3030226.
mainly refer to future stock prices, stock returns, and stock [4] C. M. Bishop, Pattern Recognition and Machine Learning. New York,
volatility. NY, USA: Springer, 2006.
Meanwhile, a forecasting model can employ different [5] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge,
U.K.: MIT Press, 2016.
methods to fuse the forecasts of base learners. The data type [6] T. Kirstein, Multidisciplinary Know-How for Smart-Textiles Developers.
of forecasts is a critical factor in selecting fusion methods. Cambridge, U.K.: Elsevier, 2013.
In general, voting and tree-based methods are the two most [7] F. Castanedo, ‘‘A review of data fusion techniques,’’ Sci. World J.,
vol. 2013, Jan. 2013, Art. no. 704504, doi: 10.1155/2013/704504.
common fusion methods for classification, while simple aver-
[8] B. V. Dasarathy, ‘‘Sensor fusion potential exploitation-innovative archi-
ages and ANN are the most commonly used fusion methods tectures and illustrative applications,’’ Proc. IEEE, vol. 85, no. 1,
for regression. pp. 24–38, Jan. 1997, doi: 10.1109/5.554206.

81376 VOLUME 10, 2022

C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

[9] T. K. Ho, J. J. Hull, and S. N. Srihari, ‘‘Decision combination in multiple [29] E. Bautu, A. Bautu, and H. Luchian, ‘‘Evolving gene expression program-
classifier systems,’’ IEEE Trans. Pattern Anal. Mach. Intell., vol. 16, ming classifiers for ensemble prediction of movements on the stock mar-
no. 1, pp. 66–75, Jan. 1994, doi: 10.1109/34.273716. ket,’’ in Proc. Int. Conf. Complex, Intell. Softw. Intensive Syst., Kraków,
[10] S. Tulyakov, S. Jaeger, V. Govindaraju, and D. Doermann, ‘‘Review Poland, Feb. 2010, pp. 108–115.
of classifier combination methods,’’ in Machine Learning in Document [30] S. Borovkova and I. Tsiamas, ‘‘An ensemble of LSTM neural networks
Analysis and Recognition, S. Marinai and H. Fujisawa, Eds. Berlin, for high-frequency stock market classification,’’ J. Forecasting, vol. 38,
Germany: Springer, 2008, pp. 361–386. no. 6, pp. 600–619, May 2019, doi: 10.1002/for.2585.
[11] M. Almasri and K. Elleithy, ‘‘Data fusion in WSNs: Architecture, tax- [31] J. Cao, Z. Li, and J. Li, ‘‘Financial time series forecasting model based
onomy, evaluation of techniques, and challenges,’’ Int. J. Sci. Eng. Res., on CEEMDAN and LSTM,’’ Phys. A, Stat. Mech. Appl., vol. 519,
vol. 6, no. 4, pp. 1620–1636, May 2015. pp. 127–139, Apr. 2019, doi: 10.1016/j.physa.2018.11.061.
[12] L. Zhang, Y. Xie, L. Xidao, and X. Zhang, ‘‘Multi-source heteroge- [32] S. Carta, A. Corriga, A. Ferreira, D. R. Recupero, and R. Saia, ‘‘A holis-
neous data fusion,’’ in Proc. Int. Conf. Artif. Intell. Big Data (ICAIBD), tic auto-configurable ensemble machine learning strategy for financial
Chengdu, China, May 2018, pp. 47–51. trading,’’ Computation, vol. 7, no. 4, p. 67, Nov. 2019, doi: 10.3390/
[13] R. T. Clemen, ‘‘Combining forecasts: A review and annotated bibliog- computation7040067.
raphy,’’ Int. J. Forecasting, vol. 5, no. 4, pp. 559–583, Jan. 1989, doi: [33] G. Chakraborty, H. Watanabe, and B. Chakraborty, ‘‘Prediction in
10.1016/0169-2070(89)90012-5. dynamic system—A divide and conquer approach,’’ in Proc. IEEE
[14] D. P. Gandhmal and K. Kumar, ‘‘Systematic analysis and review of stock Midnight-Summer Workshop Soft Comput. Ind. Appl., Espoo, Finland,
market prediction techniques,’’ Comput. Sci. Rev., vol. 34, Nov. 2019, Jun. 2005, pp. 196–201.
Art. no. 100190, doi: 10.1016/j.cosrev.2019.08.001. [34] V. Chandrasekara, C. Tilakaratne, and M. Mammadov, ‘‘An improved
[15] G. D. Pandurang and K. Kumar, ‘‘Ensemble computations on stock probabilistic neural network model for directional prediction of a stock
market: A standardized review for future directions,’’ in Proc. IEEE Int. market index,’’ Appl. Sci., vol. 9, no. 24, p. 5334, Dec. 2019, doi:
Conf. Electr., Comput. Commun. Technol. (ICECCT), Coimbatore, India, 10.3390/app9245334.
Feb. 2019, pp. 1–6. [35] Y. Chen, B. Yang, and A. Abraham, ‘‘Flexible neural trees ensemble for
[16] O. Bustos and A. Pomares-Quimbaya, ‘‘Stock market movement fore- stock index modeling,’’ Neurocomputing, vol. 70, nos. 4–6, pp. 697–703,
cast: A systematic review,’’ Expert Syst. Appl., vol. 156, Oct. 2020, Jan. 2007, doi: 10.1016/j.neucom.2006.10.005.
Art. no. 113464, doi: 10.1016/j.eswa.2020.113464. [36] L. S. Chong, K. M. Lim, and C. P. Lee, ‘‘Stock market prediction using
[17] S. Nosratabadi, A. Mosavi, P. Duan, P. Ghamisi, F. Filip, S. Band, ensemble of deep neural networks,’’ in Proc. IEEE 2nd Int. Conf. Artif.
U. Reuter, J. Gama, and A. Gandomi, ‘‘Data science in economics: Intell. Eng. Technol. (IICAIET), Kota Kinabalu, Malaysia, Sep. 2020,
Comprehensive review of advanced machine learning and deep learn- pp. 1–5.
ing methods,’’ Mathematics, vol. 8, no. 10, p. 1799, Oct. 2020, doi: [37] Z. Dai and H. Zhu, ‘‘Forecasting stock market returns by com-
10.3390/math8101799. bining sum-of-the-parts and ensemble empirical mode decomposi-
[18] I. K. Nti, A. F. Adekoya, and B. A. Weyori, ‘‘A systematic review of fun- tion,’’ Appl. Econ., vol. 52, no. 21, pp. 2309–2323, May 2020, doi:
damental and technical analysis of stock market predictions,’’ Artif. Intell. 10.1080/00036846.2019.1688244.
Rev., vol. 53, no. 4, pp. 3007–3057, Apr. 2020, doi: 10.1007/s10462-019- [38] R. T. Das, K. K. Ang, and C. Quek, ‘‘ieRSPOP: A novel incremental
09754-z. rough set-based pseudo outer-product with ensemble learning,’’ Appl.
[19] Rahul, S. Sarangi, P. Kedia, and Monika, ‘‘Analysis of various Soft Comput., vol. 46, pp. 170–186, Sep. 2016, doi: 10.1016/j.asoc.
approaches for stock market prediction,’’ J. Statist. Manage. Syst., 2016.04.015.
vol. 23, no. 2, pp. 285–293, Feb. 2020, doi: 10.1080/09720510.2020. [39] R. Dash, S. Samal, R. Dash, and R. Rautray, ‘‘An integrated TOPSIS
1724627. crow search based classifier ensemble: In application to stock index
[20] D. Nalabala and M. Nirupamabhat, ‘‘Financial predictions based price movement prediction,’’ Appl. Soft Comput., vol. 85, Dec. 2019,
on fusion models—A systematic review,’’ in Proc. Int. Conf. Art. no. 105784, doi: 10.1016/j.asoc.2019.105784.
Emerg. Smart Comput. Informat. (ESCI), Pune, India, Mar. 2021, [40] N. Deepika and M. N. Bhat, ‘‘Predicting the E-commerce companies
pp. 28–37. stock with the aid of web advertising via search engine and social media,’’
[21] A. Thakkar and K. Chaudhari, ‘‘Fusion in stock market prediction: Revue d’Intelligence Artificielle, vol. 34, no. 1, pp. 89–94, Feb. 2020, doi:
A decade survey on the necessity, recent developments, and potential 10.18280/ria.340112.
future directions,’’ Inf. Fusion, vol. 65, pp. 95–107, Jan. 2021, doi: [41] F. Giacomel, A. C. M. Pereira, and R. Galante, ‘‘Improving financial
10.1016/j.inffus.2020.08.019. time series prediction through output classification by a neural network
[22] M. H. L. B. Abdullah and V. Ganapathy, ‘‘Neural network ensemble for ensemble,’’ in Proc. Int. Conf. Database Expert Syst. Appl., Q. Chen,
financial trend prediction,’’ in Proc. TENCON Intell. Syst. Technol. New A. Hameurlain, F. Toumani, R. Wagner, and H. Decker, Eds., Valencia,
Millennium, Kuala Lumpur, Malaysia, 2000, pp. 157–161. Spain, 2015, pp. 331–338.
[23] A. Abraham and A. AuYeung, ‘‘Integrating ensemble of intelligent sys- [42] F. Giacomel, R. Galante, and A. Pereira, ‘‘An algorithmic trading
tems for modeling stock indices,’’ in Proc. 7th Int. Work-Conf. Artif. agent based on a neural network ensemble: A case of study in north
Natural Neural Netw., J. Mira and J. R. Alverez, Eds., Menorca, Spain, American and Brazilian stock markets,’’ in Proc. IEEE/WIC/ACM Int.
2003, pp. 774–781. Conf. Web Intell. Intell. Agent Technol. (WI-IAT), Singapore, Dec. 2015,
[24] A. Abraham, C. Grosan, S. Y. Han, and A. Gelbukh, ‘‘Evolutionary pp. 230–233.
multiobjective optimization approach for evolving ensemble of intel- [43] R. T. Gonzalez, C. A. Padilha, and D. A. C. Barone, ‘‘Ensem-
ligent paradigms for stock market modeling,’’ in Proc. 4th Mex. Int. ble system based on genetic algorithm for stock market forecast-
Conf. Artif. Intell., in Lecture Notes in Computer Science, A. Gelbukh, ing,’’ in Proc. IEEE Congr. Evol. Comput., Sendai, Japan, May 2015,
Á. de Albornoz, and H. Terashima-Marín, Eds., Monterrey, Mexico, pp. 3102–3108.
2005, pp. 673–681. [44] S. A. Gyamerah, P. Ngare, and D. Ikpe, ‘‘On stock market move-
[25] Z. Akhtar and M. O. Khursheed, ‘‘Stock market prediction using a hybrid ment prediction via stacking ensemble learning method,’’ in Proc. IEEE
model,’’ in Proc. 1st Symp. Mach. Learn. Metaheuristics Algorithms Conf. Comput. Intell. Financial Eng. Econ. (CIFEr), Shenzhen, China,
Appl., Trivandrum, India, 2019, pp. 75–89. May 2019, pp. 1–8.
[26] B. Alhnaity and M. Abbod, ‘‘A new hybrid financial time series prediction [45] S. S. Hasan, R. Rahman, N. Mannan, H. Khan, J. N. Moni, and
model,’’ Eng. Appl. Artif. Intell., vol. 95, Oct. 2020, Art. no. 103873, doi: R. M. Rahman, ‘‘Improved stock price prediction by integrating data
10.1016/j.engappai.2020.103873. mining algorithms and technical indicators: A case study on Dhaka stock
[27] S. Barak, A. Arjmand, and S. Ortobelli, ‘‘Fusion of multiple diverse exchange,’’ in Proc. Conf. Comput. Collective Intell. Technol. Appl.,
predictors in stock market,’’ Inf. Fusion, vol. 36, pp. 90–102, Jul. 2017, N. T. Nguyen, G. A. Papadopoulos, P. Jędrzejowicz, B. Trawiński, and
doi: 10.1016/j.inffus.2016.11.006. G. Vossen, Eds., Nicosia, Cyprus, 2017, pp. 288–297.
[28] S. Basak, S. Kar, S. Saha, L. Khaidem, and S. R. Dey, ‘‘Predict- [46] M. Jiang, J. Liu, L. Zhang, and C. Liu, ‘‘An improved stacking framework
ing the direction of stock market prices using tree-based classifiers,’’ for stock index prediction by leveraging tree-based ensemble models
North Amer. J. Econ. Finance, vol. 47, pp. 552–567, Jan. 2019, doi: and deep learning algorithms,’’ Phys. A, Stat. Mech. Appl., vol. 541,
10.1016/j.najef.2018.06.013. Mar. 2020, Art. no. 122272, doi: 10.1016/j.physa.2019.122272.

VOLUME 10, 2022 81377

 C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

[47] R. S. Joao, T. F. Guidoni, J. R. Bertini, M. D. C. Nicoletti, and [65] S. C. Nayak, C. S. K. Dash, A. K. Behera, and S. Dehuri, ‘‘Improving
A. O. Artero, ‘‘Stock closing price forecasting using ensembles of stock market prediction through linear combiners of predictive models,’’
constructive neural networks,’’ in Proc. Brazilian Conf. Intell. Syst., in Computational Intelligence in Data Mining, H. S. Behera, J. Nayak,
São Paulo, Brazil, Oct. 2014, pp. 109–114. B. Naik, and D. Pelusi, Eds. Singapore: Springer, 2020, pp. 415–426.
[48] P. Khuwaja, S. A. Khowaja, I. Khoso, and I. A. Lashari, ‘‘Prediction of [66] M. T. F. Nezhad and B. M. Bidgoli, ‘‘Development of an ensemble
stock movement using phase space reconstruction and extreme learn- learning-based intelligent model for stock market forecasting,’’ Scientia
ing machines,’’ J. Exp. Theor. Artif. Intell., vol. 32, no. 1, pp. 59–79, Iranica, vol. 28, no. 1, pp. 395–411, Sep. 2021.
Jan. 2020, doi: 10.1080/0952813x.2019.1620870. [67] H. Niu, K. Xu, and W. Wang, ‘‘A hybrid stock price index forecasting
[49] C. Krauss, X. A. Do, and N. Huck, ‘‘Deep neural networks, gradient- model based on variational mode decomposition and LSTM network,’’
boosted trees, random forests: Statistical arbitrage on the S&P 500,’’ Appl. Intell., vol. 50, pp. 4296–4309, Jul. 2020, doi: 10.1007/s10489-020-
Eur. J. Oper. Res., vol. 259, no. 2, pp. 689–702, Jun. 2017, doi: 01814-0.
10.1016/j.ejor.2016.10.031. [68] I. K. Nti, A. F. Adekoya, and B. A. Weyori, ‘‘Efficient stock-market
[50] W. Kristjanpoller R. and K. Michell V., ‘‘A stock market risk forecasting prediction using ensemble support vector machine,’’ Open Comput. Sci.,
model through integration of switching regime, ANFIS and GARCH vol. 10, no. 1, pp. 153–163, Jul. 2020, doi: 10.1515/comp-2020-0199.
techniques,’’ Appl. Soft Comput., vol. 67, pp. 106–116, Jun. 2018, doi: [69] D. K. Padhi and N. Padhy, ‘‘Prognosticate of the financial market utilizing
10.1016/j.asoc.2018.02.055. ensemble-based conglomerate model with technical indicators,’’ Evol.
[51] R. J. Kuo, L. C. Lee, and C. F. Lee, ‘‘Integration of artificial neural Intell., vol. 14, no. 2, pp. 1035–1051, Jan. 2021, doi: 10.1007/s12065-
networks and fuzzy delphi for stock market forecasting,’’ in Proc. IEEE 020-00528-z.
Int. Conf. Syst., Man Cybern. Inf. Intell. Syst., Beijing, China, Oct. 1996, [70] U. Pasupulety, A. A. Anees, S. Anmol, and B. R. Mohan, ‘‘Predicting
pp. 1073–1078. stock prices using ensemble learning and sentiment analysis,’’ in Proc.
[52] S. Lahmiri, ‘‘Ensemble with radial basis function neural networks for IEEE 2nd Int. Conf. Artif. Intell. Knowl. Eng. (AIKE), Sardinia, Italy,
Casablanca stock market returns prediction,’’ in Proc. 2nd World Conf. Jun. 2019, pp. 215–222.
Complex Syst. (WCCS), Agadir, Morocco, Nov. 2014, pp. 469–474. [71] J. Patel, S. Shah, P. Thakkar, and K. Kotecha, ‘‘Predicting stock
[53] S. Lahmiri, ‘‘A technical analysis information fusion approach for stock market index using fusion of machine learning techniques,’’
price analysis and modeling,’’ Fluctuation Noise Lett., vol. 17, no. 1, Expert Syst. Appl., vol. 42, no. 4, pp. 2162–2172, Mar. 2015, doi:
Jan. 2018, Art. no. 1850007, doi: 10.1142/s0219477518500074. 10.1016/j.eswa.2014.10.031.
[54] S. Lahmiri and M. Boukadoum, ‘‘An ensemble system based on hybrid [72] M. Pulido, P. Melin, and O. Castillo, ‘‘Particle swarm optimization of
EGARCH-ANN with different distributional assumptions to predict ensemble neural networks with fuzzy aggregation for time series predic-
S&P 500 intraday volatility,’’ Fluctuation Noise Lett., vol. 14, no. 01, tion of the Mexican stock exchange,’’ Inf. Sci., vol. 280, pp. 188–204,
Mar. 2015, Art. no. 1550001, doi: 10.1142/s0219477515500017. Oct. 2014, doi: 10.1016/j.ins.2014.05.006.
[55] S. Lahmiri and M. Boukadoum, ‘‘Intelligent ensemble forecasting sys- [73] B. Qian and K. Rasheed, ‘‘Stock market prediction with multiple
tem of stock market fluctuations based on symetric and asymetric classifiers,’’ Appl. Intell., vol. 26, no. 1, pp. 25–33, Jan. 2007, doi:
wavelet functions,’’ Fluctuation Noise Lett., vol. 14, no. 4, Nov. 2015, 10.1007/s10489-006-0001-7.
Art. no. 1550033, doi: 10.1142/s0219477515500339. [74] X. Qiu, H. Zhu, P. N. Suganthan, and G. A. J. Amaratunga, ‘‘Stock
[56] K. C. Lee and W. C. Kim, ‘‘Integration of human knowledge and machine price forecasting with empirical mode decomposition based ensemble ν-
knowledge by using fuzzy post adjustment: Its performance in stock support vector regression model,’’ in Proc. 1st Int. Conf. Comput. Intell.
market timing prediction,’’ Expert Syst., vol. 12, no. 4, pp. 331–338, Commun. Bus. Anal., Kolkata, India, 2017, pp. 22–34.
Nov. 1995, doi: 10.1111/j.1468-0394.1995.tb00270.x. [75] F. Qolipour, M. Ghasemzadeh, and N. Mohammad-Karimi, ‘‘The pre-
[57] X. Li, Z. Deng, and J. Luo, ‘‘Trading strategy design in financial dictability of tree-based machine learning algorithms in the big
investment through a turning points prediction scheme,’’ Expert data context,’’ Int. J. Eng., vol. 34, no. 1, pp. 82–89, 2021, doi:
Syst. Appl., vol. 36, no. 4, pp. 7818–7826, May 2009, doi: 10.5829/ije.2021.34.01a.10.
10.1016/j.eswa.2008.11.014. [76] H. Rezaei, H. Faaljou, and G. Mansourfar, ‘‘Stock price prediction using
[58] X.-L. Liang and W. W. Y. Ng, ‘‘Stock investment decision support using deep learning and frequency decomposition,’’ Expert Syst. Appl., vol. 169,
an ensemble of L-GEM based on RBFNN diverse trained from different May 2021, Art. no. 114332, doi: 10.1016/j.eswa.2020.114332.
years,’’ in Proc. Int. Conf. Mach. Learn. Cybern., Xi’an, China, Jul. 2012, [77] R. Sawhney, P. Mathur, A. Mangal, P. Khanna, R. R. Shah, and
pp. 394–399. R. Zimmermann, ‘‘Multimodal multi-task financial risk forecasting,’’
[59] G. Lin, A. Lin, and J. Cao, ‘‘Multidimensional KNN algorithm based in Proc. 28th ACM Int. Conf. Multimedia (MM). USA: Association for
on EEMD and complexity measures in financial time series forecast- Computing Machinery, Oct. 2020, pp. 456–465. [Online]. Available:
ing,’’ Expert Syst. Appl., vol. 168, Apr. 2021, Art. no. 114443, doi: https://www.scopus.com/record/display.uri?eid=2-s2.0-85097217724&
10.1016/j.eswa.2020.114443. doi= 10.1145%2f3394171.3413752&origin=inward&txGid=c3a29e36b2
[60] Y. Lin, Y. Yan, J. Xu, Y. Liao, and F. Ma, ‘‘Forecasting stock 6b234a84ab3f08c2ba19e4& featureToggles=FEATURE_NEW_DOC_
index price using the CEEMDAN-LSTM model,’’ North Amer. J. DETAILS_EXPORT:1,FEATURE_EXPORT_REDESIGN:0
Econ. Finance, vol. 57, Jul. 2021, Art. no. 101421, doi: 10.1016/ [78] A. Senanayake, ‘‘Automated neural-ware system for stock market pre-
j.najef.2021.101421. diction,’’ in Proc. IEEE Conf. Cybern. Intell. Syst., Singapore, Dec. 2004,
[61] G. Liu, F. Xiao, C.-T. Lin, and Z. Cao, ‘‘A fuzzy interval time- pp. 1166–1171.
series energy and financial forecasting model using network- [79] N. Sharma and A. Juneja, ‘‘Combining of random forest estimates
based multiple time-frequency spaces and the induced-ordered using LSboost for stock market index prediction,’’ in Proc. 2nd
weighted averaging aggregation operation,’’ IEEE Trans. Fuzzy Int. Conf. Converg. Technol. (I2CT), Mumbai, India, Apr. 2017,
Syst., vol. 28, no. 11, pp. 2677–2690, Nov. 2020, doi: 10.1109/tfuzz. pp. 1199–1202.
2020.2972823. [80] S. Sun, Y. Wei, and S. Wang, ‘‘AdaBoost-LSTM ensemble learning
[62] I. E. Livieris, A. Kanavos, G. Vonitsanos, N. Kiriakidou, A. Vikatos, for financial time series forecasting,’’ in Computational Science—ICCS
K. Giotopoulos, and V. Tampakas, ‘‘Performance evaluation of an SSL 2018 (Lecture Notes in Computer Science), Y. Shi, H. Fu, Y. Tian,
algorithm for forecasting the dow Jones index stocks,’’ in Proc. 9th V. V. Krzhizhanovskaya, M. H. Lees, and J. Dongarra, Eds. Wuxi, China:
Int. Conf. Inf., Intell., Syst. Appl. (IISA), Zakynthos, Greece, Jul. 2018, Springer, 2018, pp. 590–597.
pp. 1–8. [81] Z. Tang, T. Zhang, J. Wu, X. Du, and K. Chen, ‘‘Multistep-ahead
[63] P. Melin, J. Soto, O. Castillo, and J. Soria, ‘‘A new approach stock price forecasting based on secondary decomposition technique
for time series prediction using ensembles of ANFIS models,’’ and extreme learning machine optimized by the differential evolution
Expert Syst. Appl., vol. 39, no. 3, pp. 3494–3506, Feb. 2012, doi: algorithm,’’ Math. Problems Eng., vol. 2020, pp. 1–13, Jul. 2020, doi:
10.1016/j.eswa.2011.09.040. 10.1155/2020/2604915.
[64] K.-S. Moon, S. Jun, and H. Kim, ‘‘Speed up of the majority voting [82] A. Vlasenko, N. Vlasenko, O. Vynokurova, Y. Bodyanskiy, and
ensemble method for the prediction of stock price directions,’’ Econ. D. Peleshko, ‘‘A novel ensemble neuro-fuzzy model for financial time
Comput. Econ. Cybern. Stud. Res., vol. 52, no. 1, pp. 215–228, Mar. 2018, series forecasting,’’ Data, vol. 4, no. 3, p. 126, Aug. 2019, doi:
doi: 10.24818/18423264/52.1.18.13. 10.3390/data4030126.

81378 VOLUME 10, 2022

C. Zhang et al.: Decision Fusion for Stock Market Prediction: A Systematic Review

[83] L. Wang and J. Wu, ‘‘Neural network ensemble model using PPR and LS- CHENG ZHANG (Member, IEEE) received the
SVR for stock market forecasting,’’ in Advanced Intelligent Computing, bachelor’s degree from the School of Informa-
D.-S. Huang, Y. Gan, V. Bevilacqua, and J. C. Figueroa, Eds. Berlin, tion Science and Technology, Beijing Institute of
Germany: Springer, 2011, pp. 1–8. Technology, Beijing, in 2007. He is currently pur-
[84] Q. Wang, W. Xu, and H. Zheng, ‘‘Combining the wisdom of crowds suing the Ph.D. degree with the Razak Faculty
and technical analysis for financial market prediction using deep random of Technology and Informatics (RFTI), Universiti
subspace ensembles,’’ Neurocomputing, vol. 299, pp. 51–61, Jul. 2018, Teknologi Malaysia. His current research interests
doi: 10.1016/j.neucom.2018.02.095.
include time–frequency decomposition ensemble,
[85] S. M. Winkler, B. Castaño, S. Luengo, S. Schaller, G. Kronberger, and
Gaussian process regression, Bayesian optimiza-
M. Affenzeller, ‘‘Heterogeneous model ensembles for short-term predic-
tion of stock market trends,’’ Int. J. Simul. Process Model., vol. 11, no. 6, tion, decision fusion, financial market prediction,
pp. 504–513, Jan. 2016, doi: 10.1504/ijspm.2016.082914. and fintech.
[86] J. Wu, T. Zhou, and T. Li, ‘‘A hybrid approach integrating multiple
ICEEMDANs, WOA, and RVFL networks for economic and financial
time series forecasting,’’ Complexity, vol. 2020, pp. 1–17, Oct. 2020, doi:
10.1155/2020/9318308.
[87] Q. Wu, Y. Chen, and Z. Liu, ‘‘Ensemble model of intelligent paradigms
for stock market forecasting,’’ in Proc. 1st Int. Workshop Knowl. Discov-
ery Data Mining, Adelaide, SA, Australia, Jan. 2008, pp. 205–208.
[88] Q. Xie, G. Cheng, X. Xu, and Z. Zhao, ‘‘Research based on stock
predicting model of neural networks ensemble learning,’’ in Proc. 2nd
Int. Conf. Electron. Inf. Technol. Comput. Eng., Shanghai, China, 2018,
p. 02029.
[89] W. Xu, M. Zuo, M. Zhang, and R. He, ‘‘A neural network-based
ensemble forecasting method for financial market prediction,’’ Int.
J. Adv. Mech. Syst., vol. 3, no. 4, pp. 259–267, Oct. 2011, doi:
10.1504/ijamechs.2011.043374. NILAM N. A. SJARIF received the Ph.D. degree
[90] Y. Xu, C. Yang, S. Peng, and Y. Nojima, ‘‘A hybrid two-stage financial
in human action recognition behavior for video
stock forecasting algorithm based on clustering and ensemble learning,’’
surveillance from Universiti Teknologi Malaysia,
Int. J. Speech Technol., vol. 50, no. 11, pp. 3852–3867, Jul. 2020, doi:
10.1007/s10489-020-01766-5. Skudai, in 2015. She currently is a Senior Lecturer
[91] B. Yang, Z.-J. Gong, and W. Yang, ‘‘Stock market index prediction using with the Razak Faculty of Technology and Infor-
deep neural network ensemble,’’ in Proc. 36th Chin. Control Conf. (CCC), matics (RFTI), Universiti Teknologi Malaysia. Her
Dalian, China, Jul. 2017, pp. 3882–3887. research interests include machine learning, deep
[92] Y. Yujun, Y. Yimei, and X. Jianhua, ‘‘A hybrid prediction method for stock learning, pattern recognition, image processing,
price using LSTM and ensemble EMD,’’ Complexity, vol. 2020, pp. 1–16, and big data analytics. She has involved in several
Dec. 2020, doi: 10.1155/2020/6431712. research projects collaborating with industry, such
[93] X.-D. Zhang, A. Li, and R. Pan, ‘‘Stock trend prediction based on as JAKIM and Cybersecurity Malaysia.
a new status box method and AdaBoost probabilistic support vector
machine,’’ Appl. Soft Comput., vol. 49, pp. 385–398, Dec. 2016, doi:
10.1016/j.asoc.2016.08.026.
[94] H. Zheng, S. R. Kulkarni, and H. V. Poor, ‘‘A sequential predictor
retraining algorithm and its application to market prediction,’’ Ann. Oper.
Res., vol. 208, no. 1, pp. 209–225, May 2013, doi: 10.1007/s10479-013-
1396-2.
[95] Y. Zhong, Q. Zhao, and W. Rao, ‘‘Predicting stock market indexes with
world news,’’ in Proc. 4th Int. Conf. Syst. Informat. (ICSAI), Nov. 2017,
pp. 1535–1540.
[96] F. Zhou, Q. Zhang, D. Sornette, and L. Jiang, ‘‘Cascading logistic regres-
sion onto gradient boosted decision trees for forecasting and trading stock
indices,’’ Appl. Soft Comput., vol. 84, Nov. 2019, Art. no. 105747, doi:
10.1016/j.asoc.2019.105747.
[97] Z. H. Zhou, ‘‘Ensemble learning,’’ in Encyclopedia of Biometrics, S. Z. Li ROSLINA B. IBRAHIM (Member, IEEE)
and A. Jain, Eds. Boston, MA, USA: Springer, 2009, pp. 270–273. received the Ph.D. degree in information science
[98] D. Burka, C. Puppe, L. Szepesváry, and A. Tasnádi, ‘‘Voting: A machine from Universiti Kebangsaan Malaysia, Bangi,
learning approach,’’ Eur. J. Oper. Res., vol. 299, no. 3, pp. 1003–1017, in 2013. She is an Associate Professor with the
Jun. 2022, doi: 10.1016/j.ejor.2021.10.005. Razak Faculty of Technology and Informatics
[99] T. Pike and F. Vazquez-Grande, ‘‘Combining forecasts: Can machines
(RFTI), Universiti Teknologi Malaysia. She has
beat the average?’’ SSRN, p. 26, Nov. 2020. [Online]. Available:
more than ten years of experience in conducting
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3691117, doi:
10.2139/ssrn.3691117. research and has published more than 100 publi-
[100] V. Genre, G. Kenny, A. Meyler, and A. Timmermann, ‘‘Com- cations in journals, conference proceedings, and
bining expert forecasts: Can anything beat the simple average?’’ technical reports. Her research interests include
Int. J. Forecasting, vol. 29, no. 1, pp. 108–121, Jan. 2013, doi: human–computer interaction and information systems, mainly designing,
10.1016/j.ijforecast.2012.06.004. evaluating, and adopting educational games. She is the editorial board
[101] Z. Wu and N. E. Huang, ‘‘Ensemble empirical mode decomposition: member of several local and international journals and part of the technical
A noise-assisted data analysis method,’’ Adv. Adapt. Data Anal., vol. 1, program committee for conferences.
no. 1, pp. 1–41, Jan. 2009, doi: 10.1142/s1793536909000047.

VOLUME 10, 2022 81379