International Journal of Machine Learning and Cybernetics

https://doi.org/10.1007/s13042-019-01041-1

ORIGINAL ARTICLE

Study on the prediction of stock price based on the associated

network model of LSTM
Guangyu Ding1 · Liangxi Qin1

Received: 15 August 2019 / Accepted: 14 November 2019

© The Author(s) 2019

Abstract
Stock market has received widespread attention from investors. It has always been a hot spot for investors and investment
companies to grasp the change regularity of the stock market and predict its trend. Currently, there are many methods for
stock price prediction. The prediction methods can be roughly divided into two categories: statistical methods and artifi-
cial intelligence methods. Statistical methods include logistic regression model, ARCH model, etc. Artificial intelligence
methods include multi-layer perceptron, convolutional neural network, naive Bayes network, back propagation network,
single-layer LSTM, support vector machine, recurrent neural network, etc. But these studies predict only one single value.
In order to predict multiple values in one model, it need to design a model which can handle multiple inputs and produces
multiple associated output values at the same time. For this purpose, it is proposed an associated deep recurrent neural net-
work model with multiple inputs and multiple outputs based on long short-term memory network. The associated network
model can predict the opening price, the lowest price and the highest price of a stock simultaneously. The associated network
model was compared with LSTM network model and deep recurrent neural network model. The experiments show that the
accuracy of the associated model is superior to the other two models in predicting multiple values at the same time, and its
prediction accuracy is over 95%.

Keywords Deep learning · Machine learning · Long short-term memory (LSTM) · Deep recurrent neural network ·
Associated network

1 Introduction research of a country’s economic development. Therefore,

the research on the intrinsic value and prediction of the stock
Stock market has received widespread attention from inves- market has great theoretical significance and wide applica-
tors. How to grasp the changing regularity of the stock mar- tion prospects.
ket and predict the trend of stock prices has always been a The main purpose of this paper is to design a deep net-
hot spot for investors and researchers. The rise and fall of work model to predict simultaneously the opening price, the
stock prices are influenced by many factors such as politics, lowest price and the highest price of a stock on the next day
economy, society and market. For stock investors, the trend according to the historical price of the stock and other tech-
forecast of the stock market is directly related to the acqui- nical parameter data. Therefore, it is proposed an LSTM-
sition of profits. The more accurate the forecast, the more based deep recurrent neural network model to predict the
effectively it can avoid risks. For listed companies, the stock three associated values (so it is called the associated neural
price not only reflects the company’s operating conditions network model, and abbreviated as associated net model).
and future development expectations, but also an important The associated net model is compared with LSTM and
technical index for the analysis and research of the company. LSTM-based deep recurrent neural network, and verified
Stock forecasting research also plays an important role in the the feasibility of the model by comparing the accuracy of
the three models.
* Liangxi Qin The rest of this paper is organized as follows. Section 2
[email protected] introduces the research status of stock price forecasting. Sec-
tion 3 introduces the model design of the associated neu-
1
School of Computer, Electronics and Information, Guangxi ral network model. Section 4 describes the design of the
University, Nanning 530004, Guangxi, China

13
Vol.:(0123456789)
 International Journal of Machine Learning and Cybernetics

algorithm and experimental parameters. Section 5 introduces Artificial bee colony algorithm was combined with wave-
the experimental data set, the experimental results and the let transforms and recurrent neural network for stock price
analysis on the results. Section 6 concludes the paper. forecasting. Many international stock indices were simulated
for evaluation, including the Dow Jones industrial average
(DJIA), London FTSE 100 index (FTSE), Tokyo Nikkei-225
2 Related works index (Nikkei) and the Taiwan stock exchange Capitalization
Weighted Stock Index (TAIEX). The simulation results show
There are many related researches on stock price prediction. that the system has good prediction performance and can
Support vector machines was applied to build a regression be applied to real-time trading system of stock prediction.
model of historical stock data and to predict the trend of A multi-output speaker model based on RNN-LSTM was
stocks [1]. Particle swarm optimization algorithm is used to used in the field of speech recognition [14]. The experimen-
optimize the parameters of support vector machine, which tal results show that the model is better than a single speaker
can predict the stock value robustly [2]. This study improves model, and fine-tuning under the infrastructure when adding
the support vector machine method, but particle swarm opti- new output branches. Obtaining a new output model not only
mization algorithm requires a long time to calculate. LSTM reduces memory usage but also better than training a new
was combined with naive Bayesian method to extract market speaker model. A multi-input multi-output convolutional
emotion factors to improve the performance of prediction neural network model (MIMO-Net) was designed for cell
[3]. This method can be used to predict financial markets segmentation of fluorescence microscope images [15]. The
in completely different time scales with other variables. experimental results show that this method is superior to
The emotional analysis model integrated with the LSTM the most state-of-the-art deep learning based segmentation
time series learning model to obtain a robust time series method.
model for predicting the opening price of stocks, and the Inspired by the above research, considering that some
results showed that this model could improve the accuracy parameters and indicators of a stock are associated with one
of prediction [11]. Jia [12] discussed the effectiveness of another, it is necessary to design a multi-value associated
LSTM for predicting stock price, and the study showed that neural network model that can handle multiple associated
LSTM is an effective method to predict stock profits. Real- prices of the same stock and output these parameters and
time wavelet denoising was combined with LSTM network indicators at the same time. For this purpose, it is proposed
to predict the east Asian stock index, which corrected some an associated neural network model based on LSTM deep
logic defects in previous studies [13]. Compared with the recurrent network which is established by historical data and
original LSTM, this combination model is greatly improved for predicting the opening price, lowest price and highest
with high prediction accuracy and small regression error. price of the stock on the next day.
Bagging method was used to combine multiple neural net-
work method to predict Chinese stock index (including the
Shanghai composite index and Shenzhen component index) 3 Model design
[4], each neural network was trained by back propagation
method and Adam optimization algorithm, the results show 3.1 Long short‑term memory network
that the method has different accuracy for prediction of dif-
ferent stock index, but the prediction on close is unsatisfac- Long short-term memory network (LSTM) is a particular
tory. The evolutionary method was applied to predict the form of recurrent neural network (RNN), which is the gen-
change trend of stock price [5]. The deep belief network with eral term of a series of neural networks capable of process-
inherent plasticity was used to predict the stock price time ing sequential data. LSTM is a special network structure
series [6]. Convolutional neural network was applied to pre- with three “gate” structures (shown in Fig. 1). Three gates
dict the trend of stock price [7]. A forward multi-layer neural are placed in an LSTM unit, called input gate, forgetting
network model was created for future stock price prediction gate and output gate. While information enters the LSTM’s
by using a hybrid method combining technical analysis vari- network, it can be selected by rules. Only the information
ables and basic analysis variables of stock market indicators conforms to the algorithm will be left, and the information
and BP algorithm [8]. The results show that this method that does not conform will be forgotten through the forget-
has higher accuracy in predicting daily stock price than the ting gate.
technical analysis method. An effective soft computing tech- The gate allows information to be passed selectively and
nology was designed for Dhaka Stock Exchange (DSE) to Eq. 1 shows the default activation function of the LSTM net-
predict the closing price of DSE [9]. The comparison experi- work, the sigmoid function. The LSTM can add and delete
ment with artificial neural network and adaptive neural fuzzy information for neurons through the gating unit. To determine
reasoning system shows that this method is more effective. selectively whether information passes or not, it consists of a

13
International Journal of Machine Learning and Cybernetics

where ht−1 represents the output of the previous neuron

and xt is the input of the current neuron. 𝜎 is the sigmoid
function.
The input gate determines how much new information is
added to the neuron state. First, the input layer containing
the sigmoid activation function determines which informa-
tion needs to be updated, and then a tanh layer generates
candidate vectors ĉ t , an update is made to the state of the
neuron, as shown in Eq. 4

Ct = ft ∗ Ct−1 + it ∗ Ĉ t (4)

where the calculation methods of it and Ĉ t are shown in

Eqs. 5 and 6
Fig. 1  LSTM unit structure ( [ ] )
it = 𝜎 Wi ⋅ ht−1 , xt + bi (5)

Sigmoid neural network layer and a pair multiplication opera- ( [ ] )

tion. Each element output by the Sigmoid layer is a real num- Ĉ t = tanh Wc ⋅ ht−1 , xt + bc (6)
ber between [0, 1], representing the weight through which
The output gate is used to control how many current neu-
the corresponding information passes. In the LSTM neural
ral unites state are filtered and how many controlling units
network, there is also a layer containing tanh activation func-
state are filtered which are shown in Eqs. 7 and 8
tion which shown in Eq. 2. It is used for updating the state of
( [ ] )
neurons ot = 𝜎 Wo ⋅ ht−1 , xt + bo (7)
1
𝜎(x) = (1) ( )
1 + e−x ht = Ot ∗ tanh Ct (8)

ex − e−x
tanh (x) = (2)
ex + e−x
3.2 Deep recurrent neural network
The forgetting gate of the LSTM neural network determines
what information needs to be discarded, which reads ht−1 and A LSTM-based deep recurrent neural network (DRNN)
xt, gives the neuron state Ct−1 a value of 0–1. Equation 3 shows is a variant of the recurrent neural network. To enhance
the calculation method of forgetting probability the expressive power of the model, the loop body at each
( [ ] ) moment can be repeated many times. As shown in Fig. 2, the
ft = 𝜎 Wf ⋅ ht−1 , xt + bf (3)
structure diagram of deep recurrent neural network is given.

Fig. 2  The neural network model without and with dropout

13
 International Journal of Machine Learning and Cybernetics

Deep recurrent neural network is composed of LSTM, Input layer

so its operation mechanism is same as LSTM. During the
process of constructing the task model, the dropout method LSTM
was used. Dropout refers to the temporary discarding of the dropout LSTM
neural network unit from the network according to a certain
LSTM LSTM
probability during the training of the deep learning network,
which is a means to prevent over-fitting. The principle of dropout
LSTM
dropout operation is that the neurons in each layer are ran- Output
domly deleted with probability P in a training iteration, and layer dropout
the data in this iteration are trained with the network com- LSTM
LSTM
posed of the remaining (1 − p)*N neurons, thus alleviating L1
dropout dropout
the over-fitting problem. The neural network model without
Output
dropout is shown in Fig. 2a, b is the neural network model layer LSTM
with dropout. dropout
The LSTM-based deep recurrent neural network model L2
Output
with dropout layer was used as the contrast model to verify layer
the feasibility and applicability of the proposed associated L3
neural network model. The structure of LSTM-based deep
recurrent neural network is shown in Fig. 3. Fig. 4  A structural model of associated net

3.3 Associated neural network model

price and the highest price respectively. In the Chinese stock
Since the daily opening price, the lowest price and the high- market, the maximum fluctuation of stock price is only 10%.
est price of the stock are associated to one another, and the Therefor the model recombines the output of the left branch
opening price, the lowest price and the highest price are (opening price) and the output of the LSTM network of the
respectively predicted by different networks generally, the second branch as the input parameter data of the predicted
associations between one another are separated. Therefore, lowest price, and the highest price is subject to the open-
based on the deep recurrent neural network, a structural ing price of the day, the impact of the lowest price, so the
model of multi-value associated neural network (associated output of the left branch (opening price) and the output of
net) based on LSTM is designed, it is shown in Fig. 4. the intermediate branch (lowest price) and the output of the
The specific data processing flow of the multi-value LSTM network of the third branch form the highest of the
associated neural network model is shown in Fig. 5. Data new data forecast price.
through the input layer to all three branches simultaneously. In the model training phase, the total loss ­Ltotal is used as
These three branches predict the opening price, the lowest the evaluation function, and the goal is to get the minimal
value. The calculation method of the total loss is shown in
Eq. 9
Fig. 3  Deep recurrent neural
network based on LSTM Input layer n
1∑
Ltotal = L (9)
n i=1 i

LSTM

dropout
4 Design of algorithm and experiments

Regression method is used to predict a specific value,

LSTM which is not a pre-defined category, but an arbitrary real
number. Regression problem generally has only one output,
and the output is the predicting value. The loss function
dropout used in regression problems commonly is the mean square
error (MSE) (Eq. 10). It is the expectation of the square
Output of the difference between the estimated parameter and the
layer actual parameter. MSE can evaluate the degree of change of
the function. The smaller the value of MSE, the better the

13
International Journal of Machine Learning and Cybernetics

Basic input
(7 input nodes)

Forecast the lowest

price of DRNN
network
(7 input nodesˈ1 output
node) Forecast the lowest price
OP of DRNN network
(8 input nodesˈ1 output node)

Forecast current Forecast the highest price

OP of DRNN network
opening price
(OP) Forecast current LP (9 input nodes, 1 output node)
lowest price (LP) Forecast the
highest price(HP)

Fig. 5  The data processing flow chart of associated net

accuracy of the prediction model describing the experimen- iterations in the model, the training will continue to reduce
tal data. Therefore, in the training phase, MSE is used as the the total loss, otherwise training will stop.
criterion to measure the quality of a network model
∑n � �2
� i=1
yi − y�i
(10) 4.2 Parameter setting
MSE (y, y ) =
n
There is a parameter of step size in the input of the LSTM
neural network that means how many historical data to
4.1 Algorithm remember as a reference for predicting the current price.
In order to use a relatively good step size in the experiment
Deep learning often requires a lot of time and computational of the multi-value associated model, a comparison experi-
resources to train. It is needed to find an optimization algo- ment is performed with 6112 sample data, at the step size
rithm that requires less resources and has faster convergence of 5, 10, 20 and 30, and with the iteration number of 50.
speed. The Adam optimization algorithm is an extension The loss variation graphs are shown in Figs. 7, 8, 9 and
of the stochastic gradient descent algorithm and has great 10, separately.
advantages in solving the non-convex optimization problem. According to the loss variation graph at the step size
During the training phase, the Adam optimization of 5, 10, 20 and 30, it is found that the loss at the step
algorithm is used in the model, and ­Ltotal is used as the size of 10 and 20 decreases the fastest and finally reaches
evaluation function. Multiple values associated with neural a steady state. By comparing the average loss as shown
network model algorithm framework as shown in Fig. 6, in Table 1, it is found that the average loss at step size of
the first input sequence data to Associated Net model, it five is the lowest. The average loss at the step size of 20
contains three DRNN networks in Associated Net model. differs from the loss at the step size by 0.0014901(shown
Each DRNN network produces a loss, and the losses sum in Table 1). Considering the loss variation graph and the
of these three DRNN networks is the total loss. Then the average loss comprehensively, 20 is chosen as the step size
Adam algorithm is used to optimize the total loss. When in the model.
the number of iterations did not reach the set number of

Fig. 6  The framework of associ-

ated net algorithm

13
 International Journal of Machine Learning and Cybernetics

Fig. 7  Loss variation graph at the step size of 5 Fig. 10  Loss variation graph at the step size of 30

Table 1  Average variance loss of different steps

Step 5 10 20 30

Loss 0.0148642 0.0181064 0.0163543 0.0179096

5 Experimental results and analysis

5.1 Dataset

The experimental data in this paper are the actual historical

data downloaded from the Internet. Three data sets were
used in the experiments, one is Shanghai composite index
000001 and the others are two stocks of PetroChina (stock
code 601857) on Shanghai stock exchange and ZTE (stock
Fig. 8  Loss variation graph at the step size of 10 code 000063) on Shenzhen stock exchange. Shanghai com-
posite index has 6112 historical data; PetroChina has 2688
historical data and ZTE has 4930 historical data. Each data
set is divided into a training set and a test set in chronologi-
cal order at the ratio of 4:1. Each data set has seven technical
parameters. It is used these technical parameters as basic
input attributes, and the OP, LP and HP of the next day
as output values of the model. The identifiers of technical
parameters related to stock are shown in Table 2.
Due to the different measurement unit of different stock
index data, for avoiding the impact of different measure-
ment unit, all the attribute data are normalized to fall within
a same range. In this paper, the min–max normalization
method is used. The normalization function is shown in
Eq. 11
x − min
x� = (11)
max − min
Fig. 9  Loss variation graph at the step size of 20

13
International Journal of Machine Learning and Cybernetics

Table 2  The identifiers used Parameter name Identifier square error of Associated Net is larger than the average
for stock related technical mean square error of LSTM and DRNN. Because our model
parameters Open price OP is more complex and requires a larger number of iterations.
Close price CP In order to verify this conjecture, several experiments
Lowest price LP is conducted on Associated Net. The opening price, the
Highest price HP highest price and the lowest price of the next day were
Volumes V trained and predicted by the Associated Net model. As
Money M shown in Table 4, and the experimental results proved our
Change C conjecture. The root of this problem is that the associated
network model is composed of multiple deep-recurrent
neural networks. The model is complex, the number of
Table 3  Average variance loss of three models under different train- neurons is large, and multiple output losses are combined,
ing times so the loss of the model decreases slowly. According to the
Training times LSTM DRNN Associated net analysis experiment, the model loss chart of each model
for 200 iterations is drawn, as shown in Figs. 11, 12, and
50 0.0377711 0.0152948 0.037064
13. The output of the Associated Net is the total loss and
100 0.0147428 0.0191181 0.029533
his three sub-losses (opening price loss, lowest price loss,
200 0.0132838 0.00721601 0.026126
highest price loss). From the analysis for the loss chart, it
300 0.00418345 0.0104519 0.019745
is found that the loss of each model is gradually decreas-
500 0.00818345 0.0106546 0.014983
ing. The LSTM model has multiple fluctuations during
the training process. DRNN and Associated Net are very
stable. Moreover, the individual sub-loss of the associated
Through the normalization operation, the data is scaled network model is also gradually decreased. As shown in
to [0, 1], which not only speeds up the gradient descent to Table 4, although the total loss of Associated Net is higher
find the optimal solution, but also improves the accuracy. than that of the other two models, its sub-loss is very low,

5.2 Experimental analysis of training phase

Using the training data set of the Shanghai Stock Index,

Associated Net is compared with LSTM network and
LSTM-based deep-recurrent neural network (DRNN)
in the experiments. The highest price of the stock of the
next day was trained and predicted respectively by LSTM,
DRNN and Associated Net. As shown in Table 3, it is found
that the mean square error of the three models gradually
decreased with the increase of training times. LSTM net-
work and DRNN had experienced a slight fluctuation. From
the dimension analysis of the same training times, with
the increase of the training times, the LSTM average mean
square error is lower than the other two models, but in the
test phase, the LSTM has the worst prediction effect and
the lowest average accuracy. Because LSTM has been over-
Fig. 11  Loss of LSTM
fitting as the number of training increases. The average mean

Table 4  The average square loss Times Average square loss of Average square loss of Average square loss of Average loss
of Associated Net open price lowest price highest price of three losses

50 0.0460407 0.0341459 0.0310071 0.037064

100 0.0345082 0.031234 0.0228587 0.029533
200 0.0265043 0.028083 0.0237893 0.026126
300 0.023574 0.0184222 0.0172387 0.019745
500 0.015052 0.0145609 0.0153359 0.014983

13
 International Journal of Machine Learning and Cybernetics

Fig. 14  Loss variation of ZTE stock training model

Fig. 12  Loss of DRNN

Fig. 13  Loss of associated net Fig. 15  Loss variation of PetroChina stock training model

and by increasing the number of iterations, the total loss • PetroChina has a large circulation, and the stock price
of Associated Net is gradually reduced. fluctuation is relatively small, so that a good fitting effect
In order to verify the universality of the model, the his- can be obtained quickly. ZTE’s stock price fluctuations
torical data of two stocks of PetroChina and ZTE are used are relatively larger, so that more training data is needed
to verify the universality of the model. The experimental to obtain a good fitting effect.
results are shown in Figs. 14 and 15. Combined Fig. 13 with
Table 5, it is concluded that the model fits PetroChina data 5.3 Experimental analysis in the test phase
better. The data fitting result of ZTE is relatively poor at the
beginning, but it gradually becomes better; In the end, their In order to verify the training of each model in the training
average loss of mean square error became similar. Through phase, the three models were tested separately using a test
the experiments, it is found that the more the training data, set of multiple stocks. The mean square error (MSE) is the
the better the model fitting effect. Further more, while the expected value of the square of the difference between the
number of iterations of the model training was increased estimated value of the parameter and the true value of the
appropriately, and the loss of the model decreased gradually. parameter, MAE is the average of the absolute error, and
The above results are due to the following reasons. MAE can better reflect the real situation of the predicted
value error. Therefor in the test, the average absolute error
• The model is complex and needs large amount of data to of MAE (mean absolute error) (Eq. 12) was used as the
train the parameters of each neuron. evaluation index to calculate the degree of deviation, and
the result of 1 − MAE was used as the average accuracy of

13
International Journal of Machine Learning and Cybernetics

Table 5  Average square loss Stock Open price Lowest price Highest price Average of
of different data sets in the three losses
Associated Net model
PetroChina 0.036184 0.032914 0.034226 0.034444
ZTE 0.032247 0.037305 0.031708 0.033753
Shanghai Index 0.023419 0.021976 0.030776 0.02539

Table 6  Average accuracy of three models

Model LSTM DRNN Associated Net
Open Lowest Highest
price price price

Average 0.787925 0.973754 0.971999 0.956185 0.974434

accuracy

Table 7  Average accuracy of different data sets on the associated net

model

Stock Open price Lowest price Highest price

PetroChina 0.986331 0.9795 0.981839

ZTE 0.963979 0.956655 0.96012
Shanghai Index 0.971999 0.956185 0.974434
Fig. 16  The test result of the LSTM model

the model and the measurement index. Average accuracy of

the three models was shown in Table 6, and average accu-
racy of Associated Net model with different data sets was
shown in Table 7
n
1 ∑|
MAE = f − yi || (12)
n i=1 | i

There are large errors between LSTM prediction values

and the real data (shown in Fig. 16); Fig. 17 shows the com-
parison of DRNN prediction values and real data. The pre-
diction values and the real data are almost coincident, and
the deviation between the prediction values and the real data
is much small, indicating that the performance of DRNN is
better than LSTM in the test data. The deviation between the
three prediction results of Associated Net and the real data Fig. 17  The test result of the DRNN model
is also small as shown in Fig. 18. From Figs. 18 and 17, it is
found that for the highest price prediction, Associated Net Net model can handle such problems very well, and it per-
fits the curve of real data better than DRNN, and the data forms better than DRNN model.
deviation of Associated Net is smaller than that of DRNN. The trained Associated Net model of PetroChina and ZTE
The comparison of the average accuracy for the three mod- were tested with the test data sets of the two stocks. The test
els is shown in Table 6. It can be found that for predicting results are shown in Figs. 19 and 20. Combined with Fig. 18,
the highest price, Associated Net model has higher average it is found that it fits well in the three data sets. From the data
accuracy than the other two models. This phenomenon con- in Table 7, it is found that ZTE’s average prediction accuracy
firms that the highest value of the next day is not only related is not as good as the other two stocks. However, the accuracy
to historical data, but also related to the opening price and of the three models is above 95%, and the test results are in
the lowest price of the same day. Therefore, the Associated line with the prior conjectures. Therefore, Associated Net

13
 International Journal of Machine Learning and Cybernetics

model can predict multiple associated values at the same

time, and the difference between the predicted value and the
real value is small.

6 Conclusion

In this paper, a multi-value associated network model of

LSTM-based deep-recurrent neural network (Associated
Net) is proposed to predict multiple prices of a stock simul-
taneously. The model structure, the algorithm framework
and the experiment design are presented. The feasibility and
accuracy of the Associated Net are verified by comparing
the model with LSTM network model and the LSTM deep-
recurrent neural network model. Multiple data sets were
Fig. 18  The test result of the Shanghai Index
used to verify the applicability of Associated Net model.
Experiments show that the average accuracy of Associated
Net model is not only better than that of the other two mod-
els. Moreover, it can predict multiple values simultaneously,
and the average accuracy of each predicted value is over
95%. Although the model achieves good effect, there are
still some aspects can be improved. For example, simple
arithmetic mean algorithm is used in the calculation of total
loss in training phase, and the goal is to optimize the model
by reducing the total loss. This loss calculation method does
not take into count the relationship between each sub-loss,
as well as some details when the total loss is the minimum,
such as the extreme situation of each sub-loss and the oscil-
lation in the process of loss reduction. In the next step, we
will study the dimension reduction of the input parameters
and the optimizing the loss calculation method to improve
the average accuracy of the model.

Fig. 19  The test result of the PetroChina in associated net model Acknowledgements This work is partially supported by the Science
and Technology Project of Guangxi(Guike AB16380260) and Spe-
cialized Scientific Research in Public Welfare Industry (Meteorology)
(GYHY201406027).

Open Access This article is distributed under the terms of the Crea-
tive Commons Attribution 4.0 International License (http://creat​iveco​
mmons​.org/licen​ses/by/4.0/), which permits unrestricted use, distribu-
tion, and reproduction in any medium, provided you give appropriate
credit to the original author(s) and the source, provide a link to the
Creative Commons license, and indicate if changes were made.

References
1. Xia Y, Liu Y, Chen Z (2013) Support vector regression for predic-
tion of stock trend. 6th Int Conf Inf Manag Innov Manag Ind Eng
(ICIII) 2:123–126
2. Sands TM, Tayal D, Morris ME, Monteiro ST (2015) Robust
stock value prediction using support vector machines with parti-
cle swarm optimization. Congress on Evolutionary Computation
(CEC), pp 3327–3331
Fig. 20  The test result of the ZTE in associated net model

13
International Journal of Machine Learning and Cybernetics

3. Li J, Bu H, Wu J (2017) Sentiment-aware stock market prediction: integrated system based on artificial bee colony algorithm. Appl
a deep learning method. In: IEEE: 2017 international conference Soft Comput J 2(11):2510–2525
on service systems and service management, pp 1–6 11. Zhuge Q, Xu L, Zhang G (2017) LSTM neural network with
4. Yang B, Gong Z-J, Yang W (2017) Stock market index prediction emotional analysis for prediction of stock price. Eng Lett
using deep neural network ensemble. In: 36th Chinese Control 2(25):167–175
Conference (CCC), pp 3882–3887 12. Jia H (2016) Investigation into the effectiveness of long short
5. Tsai Y-C, Hong C-Y (2017) The application of evolutionary term memory networks for stock price prediction. arXiv [cs.NE],
approach for stock trend awareness. In: IEEE: IEEE 8th interna- pp 1–6
tional conference on awareness science and technology (iCAST), 13. Li Z, Tam V (2017) Combining the real-time wavelet denoising
pp 306–311 and long- short-term-memory neural network for predicting stock
6. Li X, Yang L, Xue F, Zhou H (2017) Time series prediction of indexes. In: IEEE: 2017 IEEE symposium series on computational
stock price using deep belief networks with Intrinsic plasticity. intelligence (SSCI), pp 1–8
In: IEEE: 2017 29th Chinese Control and Decision Conference 14. Pascual S, Bonafonte A (2016) Multi-output RNN-LSTM for
(CCDC), pp 1237–1242 multiple speaker speech synthesis and adaptation. In: IEEE:
7. Gudelek MU, Boluk SA, Ozbayoglu M (2017) A deep learning 2016 24th European Signal Processing Conference (EUSIPCO),
based stock trading model with 2-D CNN trend detection. In: pp 2325–2329
IEEE: 2017 IEEE symposium series on computational intelligence 15. Ahmed Raza SE, Cheung L, Epstein D et al (2017) MIMO-NET:
(SSCI), pp 1–8 a multi-input multi-output convolutional neural network for cell
8. Adebiyi AA, Ayo KC, Adebiyi MO, Otokiti SO (2012) Stock price segmentation in fluorescence microscopy images. In: IEEE: 2017
prediction using neural network with hybridized market indica- IEEE 14th international symposium on biomedical imaging (ISBI
tors. J Emerg Trends Comput Inf Sci 1(3):1–9 2017), pp 337–340
9. Billah M, Waheed S, Hanifa A (2015) Predicting closing stock
price using artificial neural network and adaptive neuro fuzzy Publisher’s Note Springer Nature remains neutral with regard to
inference system: the case of the dhaka exchange. Int J Comput jurisdictional claims in published maps and institutional affiliations.
Appl 11(129):975–8887
10. Hsieh TJ, Hsiao HF, Yeh WC (2011) Forecasting stock mar-
kets using wavelet transforms and recurrent neural networks: an

13