https://doi.org/10.31449/inf.v48i12.6134                                                             Informatica 48 (2024) 33–40        33 
Predicting Daily Tourist Flow in Scenic Areas Using LSTM and Big 
Data from Baidu Index 
Feng Li 
School of Hotel Management, Qingdao Vocational and Technical College of Hotel Management, Qingdao, Shandong 
266100, China 
Email: lf_feng@outlook.com 
Keywords: tourism, daily tourist flow, long short-term memory, prediction 
Received: April 28, 2024 
The accurate prediction of tourist flow in scenic areas is crucial for effective tourist area management. 
This paper introduced the Baidu index associated with search engine data into relevant indicators that 
can be used to forecast daily tourist flow in scenic spots. The long short-term memory (LSTM) algorithm 
was employed for daily tourist flow prediction. Simulation experiments were conducted on the Longkou 
Nanshan Scenic Spot in Longkou City, Shandong Province, China. The experiment first verified the 
effectiveness of the feature indicators and then compared the predictive performance of the support vector 
machine (SVM), back-propagation neural network (BPNN), and LSTM models under the situation with 
or without the Baidu index. It was found that nine feature indicators exhibited significant correlations 
with daily flow, including date type, week type, month, number of holiday days, average tourist flow, 
antecedent daily flow, weather condition, standard deviation of daily flow, and Baidu index. The LSTM 
model demonstrated higher accuracy than SVM and traditional BPNN models. When using the same 
feature indicators, the p values between the LSTM model and the other two models were both 0.001, 
indicating significant differences. Furthermore, including the Baidu index as a feature significantly 
enhanced the accuracy of the prediction algorithm. When comparing within the same prediction algorithm, 
the p value between the algorithm using the Baidu index feature and the one without the Baidu index 
feature was less than 0.05, indicating a significant difference. 
Povzetek: V raziskavi so za  napovedovanje dnevnega turističnega toka uporabili podatke iz Baidu 
indeksa.  LSTM algoritem je dosegel višjo točnost napovedi kot SVM in BPNN modeli, vključitev Baidu 
pa je bistveno izboljšala natančnost napovednega algoritma. 
 
1 Introduction 
Accurate tourist flow prediction is vital for managers of 
tourist areas to formulate effective resource allocation and 
operational strategies [1], thereby enhancing tourist 
satisfaction and safeguarding visitor safety [2]. The 
conventional approach to forecasting tourist flow is linear 
prediction. This method employs historical flow data and 
other quantifiable influencing factors for conducting 
linear regression analysis and leverages the derived linear 
pattern to forecast future flow in tourist areas. However, 
tourist flow in tourist attractions often exhibits complex 
dynamic characteristics [3], typically demonstrating 
nonlinear characteristics rather than linear ones, especially 
in specific periods or regions. As computer performance 
has advanced, machine learning algorithms, which 
simulate human brain thinking, mine the nonlinear 
patterns of tourist flow in tourist areas to realize nonlinear 
prediction. The relevant studies are reviewed in Table 1.  
 
Table 1: Related studies 
Literatur
e 
Autho
r 
Method Result 
[4] Guo The author 
proposed three 
types of transit-
oriented 
development-
based urban 
public 
transportation 
hubs according 
to various 
elements of the 
BRT 
transportation 
station. 
The author 
designed a 
method and 
process for 
predicting hub 
passenger 
flow on the 
basis of 
studying the 
current scale 
and 
characteristics 
of hub 
passenger 
flow.   
[5] Li et 
al. 
A nearest 
neighbor 
nonparametric 
The case 
study 
demonstrated 

34   Informatica 48 (2024) 33–40                                                                                                   F. Li 
regression 
method for 
short-term 
passenger flow 
prediction 
that the 
predicted 
values of the 
algorithm had 
a good fit with 
the measured 
values. 
[6] Marie
-
Sainte 
et al. 
The author 
proposed to 
explore the 
firefly 
algorithm (FA) 
and particle 
swarm 
optimization 
(PSO) for 
finding the 
optimal values 
of coefficients 
in linear 
regression (LR) 
and used LR to 
predict air 
travel demand. 
The results 
indicated that 
the LR 
prediction 
based on PSO 
performed the 
best, with a 
lower error 
rate compared 
to the LR 
based on FA 
and the LR 
alone. 
2 Daily tourist flow prediction based 
on big data 
The conventional method for predicting daily tourist flow 
at tourist attractions involves utilizing historical data on 
tourist flow and other quantifiable factors that can affect 
tourist flow to conduct regression analysis and using the 
linear regression equation obtained through fitting to 
predict daily tourist flow [7]. However, in practical 
scenarios, the fluctuation of tourist flow often exhibits 
nonlinear characteristics, and the requirement for a 
substantial amount of relevant data to ensure accuracy 
makes the fitting process challenging. As computer 
performance has advanced, machine learning algorithms 
have progressively found applications across various 
domains, including predicting tourist flow in tourist 
attractions [8]. The parallel processing capabilities of 
machine learning algorithms enable effective handling of 
big data, facilitating the extraction of nonlinear patterns. 
The variation in daily tourist flow at tourist attractions 
can be viewed as a type of time series data, where past data 
influence present data. Consequently, a recurrent neural 
network (RNN) is employed to analyze time series data. 
However, RNN encounters challenges such as gradient 
explosion or gradient disappearance when dealing with 
long data sequences [9]. To address these issues, gated 
structures are introduced as memory units within the RNN 
architecture, forming LSTM. 
To improve the accuracy of daily tourist flow 
prediction using LSTM, in addition to utilizing the flow at 
past time points as input features, other factors that can 
influence tourist flow will also be incorporated as input 
features [10]. This paper incorporated weather 
information and Internet search data for attractions as 
feature factors in predicting daily tourist flow. Weather 
conditions can influence tourists' travel; typically, better 
weather conditions make travel easier. The Internet search 
data for attractions serves a dual purpose: firstly, it reflects 
the overall attention of tourists towards attractions; 
secondly, the attraction data provided by search engines 
can guide tourists in their decision-making process. 
 
Collect data
Feature 
extraction
Training set
Test set
 Input into LSTM 
for forward 
calculation
Terminate 
training?
Finish training
The assessment 
result of the 
prediction algorithm
No, adjust LSTM parameters
Yes
 
Figure 1: LSTM-based daily tourist flow prediction 
algorithm. 
Table 2: Data features related to the prediction of daily 
tourist flow in tourist attractions 
Feature name Definition and 
representation 
Date type 
t
s 
A categorical variable, 
categorized as weekdays, 
weekends, and holidays, 
denoted by 0, 1, and 2, 
respectively 
Week type 
t
w 
A categorical variable, 
categorized into Monday 
to Sunday, represented by 
0 to 6, respectively 
Number of holiday days 
t
h 
A numerical variable, 
depending on the nature of 
the date 
Average tourist flow 
s
y 
A numerical variable 
representing the average 
daily flow for each date 
type 
Month 
t
m 
A categorical variable, 
categorized into January 
to December, represented 
by 1 to 12, respectively 
Antecedent daily tourist 
flow 
k t
y
−
 
A numerical variable, i.e., 
historical daily tourist 
flow 
Weather condition 
t
a 
A categorical variable, 
categorized into sunny, 
cloudy, rainy, snowy, and 
sandy, denoted by 0, 1, 2, 
and 3, respectively 
The standard deviation of 
tourist flow 
s
v 
A numerical variable, 
representing the standard 
deviation of daily tourist 
flow for each date type 
Baidu index 
t
se 
A numerical variable 
reflecting the extent to 
which tourists search for 
scenic spot-related 
content online. 
 

Predicting Daily Tourist Flow in Scenic Areas Using LSTM…                                     Informatica 48 (2024) 33–40   35 
The procedure of the LSTM-based daily tourist flow 
prediction is shown in Figure 1. 
① Data related to tourist attractions are collected, and 
the type of data required is shown in Table 2. 
② Feature extraction is performed on the collected 
data. The corresponding feature types selected through 
looking up relevant literature and interviews with 
professionals are shown in Table 1. A day at a tourist 
attraction is regarded as one sample, and then the feature 
in the t -th sample 
t
x is denoted as 
 
t s t k t t s t t t
se v a y m y h w s , , , , , , , ,
−
, where 
s
y and
 
s
v 
depend on 
t
s and the number of 
k t
y
−
 depends on the 
number of k required. 
③ The training and testing sets were set up. 
④ The features of the samples in the training set are 
input into the LSTM for forward computation [11], which 
is formulated as: 











 =
+ =
 +  =
+ =
+ =
+  =
−
−
−
−
−
) tanh(
) ] , [ (
~
) ] , [ tanh(
~
) ] , [ (
) ] , [ (
1
1
1
1
1
t t t
o t t o t
t t t t
C t t C t
i t t i t
f t t f t
C o h
b x h o
C i C f C
b x h C
b x h i
b x h f
 

 
 
, (1) 
where 
t
C
~
 
and
 
t
C are the temporary state and updated 
state of the memory unit at the current moment [12], 
respectively, 
t
h is the hidden state of the sequence data at 
the present moment, 
t
x is the input at the present moment, 
t
f , 
t
i
 
and
 
t
o are the outputs of the three gated units, 
namely, forgetting, input, and output, at the current 
moment, respectively, 
f
 ,
 
i
 , and
 
i

 
are weights in the 
corresponding gated unit, respectively, 
f
b ,
 
i
b and
 
o
b
 
are biases in the corresponding gated unit, respectively. 
⑤ Whether the training is finished is determined. If 
not, then the parameters in LSTM are adjusted reversely, 
and then return to step ④; if it does, stop the training and 
obtain the trained prediction algorithm. 
⑥ The performance of the prediction algorithm is 
tested by feeding the samples in the test set into the trained 
algorithm. 
3 Simulation experiment 
3.1 Study area 
This study focuses on the Longkou Nanshan Scenic Spot 
in Longkou City, Shandong Province, China. It is a 
national 5A-level tourist attraction located in Longkou 
City, a coastal city on the Jiaodong Peninsula. This 
expansive scenic area is divided into multiple zones, 
including the Religious and Cultural Park, Historical and 
Cultural Park, East Sea Tourism Resort, etc. The Religious 
and Cultural Park serves as the core of the Nanshan Scenic 
Spot and consists of various temples. The Historical and 
Cultural Park was meticulously designed to showcase 
different dynasties in chronological order. The coastline of 
the East Sea Tourism Resort stretches for 20 kilometers. It 
is divided into various areas including the seaside tourist 
zone, wellness and leisure zone, villa residential zone, 
commercial and service zone, as well as cultural and 
educational zone. It is a comprehensive tourism resort that 
integrates living, tourism, leisure, and humanistic 
education with high technological content. The resort 
boasts a pristine ecological environment where people 
coexist harmoniously with nature. 
3.2 Experimental data 
The data required for the simulation experiments are 
outlined in Table 1 in the preceding section. Specifically, 
the tourist flow data for the Longkou Nanshan Scenic Spot 
was sourced from the Longkou City People's Government 
website. The chosen period for this data was from January 
2017 to December 2022. Meteorological data was 
acquired from the National Meteorological Center website 
(nmc.gov.cn/index.html), spanning the same time frame. 
Date and month information was annotated according to 
the calendar for the specified time range, and holidays 
were designated based on official announcements during 
the selected period. Baidu index data related to this scenic 
spot was collected from the Baidu website, employing 
keywords in the same time period. Samples from January 
2017 to December 2020 constituted the training set, while 
samples from January 2021 to December 2022 formed the 
test set. The values missing in the collected data were 
supplemented using the interpolation method. 
When using the Baidu website to obtain the Baidu 
index related to scenic spots [13], the selection of 
keywords is essential. The processing steps in this study 
are as follows. 
① A keyword thesaurus was established. “Longkou 
Nanshan Scenic Spot” and the name of its neighboring 
scenic spots were used as the root keywords, and then the 
beginning and end of these root keywords were expanded. 
The expansion of the beginning of the word means adding 
the name of the superior place, and the expansion of the 
end of the word means adding the vocabulary related to 
tourism elements. Taking "Longkou Nanshan Scenic 
Spot" as an example, the word beginning can be expanded 
to "Longkou City Longkou Nanshan Scenic Spot", and the 
word end can be expanded to "Longkou Nanshan Scenic 
Spot special food". ② The Baidu index of the keyword in 
the Baidu website, i.e., the overall trend of the search 
volume of the keyword, was crawled using a crawler 
program. The Baidu index was arranged in the order of the 
day. ③ The keywords with too low search volume and 
their corresponding Baidu index were deleted. 
3.3 Experimental setup 
The parameters of the LSTM algorithm obtained through 
the orthogonal experiment for predicting daily tourist flow 
in scenic spots are presented in Table 3. Additionally, 
experiments were also conducted on both support vector 
machine (SVM) and traditional back-propagation neural 

36   Informatica 48 (2024) 33–40                                                                                                   F. Li 
network (BPNN) algorithms to assess the LSTM 
algorithm's performance. The relevant parameters for the 
SVM algorithm are as follows: the kernel function utilized 
the sigmoid function [14], and the penalty parameter was 
set to 1. As for the traditional BPNN algorithm, the 
parameters were configured as follows: the number of 
nodes in the input and output layers aligned with the 
LSTM algorithm, the hidden layer was set to one layer 
with 128 nodes, the activation function adopted sigmoid, 
and the learning rate and the number of training iterations 
were consistent with those of the LSTM algorithm. 
Table 3: Parameter settings of LSTM. 
Parameter Setting Parameter Setting 
Input layer Nine 
nodes 
Hidden 
layer 
Three 
layers with 
64 nodes 
per layer 
Hidden 
layer 
activation 
function 
Sigmoid Output 
layer 
One node 
Learning 
rate 
0.02 Maximum 
number of 
training 
sessions 
500 
Batch size 100 Optimizer Stochastic 
gradient 
descent 
 
In addition to the aforementioned experiments, robustness 
tests were also conducted on the LSTM algorithm. In these 
tests, data samples from Longkou People's Park and 
Huangshui River Wetland Park in Longkou City were 
collected from January 2021 to December 2022 using the 
method described earlier. Then, the performance of the 
LSTM algorithm in other datasets was evaluated by 
applying it to predict data based on these data samples. 
3.4 Evaluation indicators 
For the performance of the three prediction algorithms 
tested in this paper, the root-mean-square error (RMSE) 
and mean absolute percentage error (MAPE) [15] was 
used to measure the performance, which is formulated as: 








−
=
−
=


=
% 100
ˆ
1
) ˆ (
1
2
N
i i
i i
i
i i
y
y y
N
MAPE
N
y y
RMSE
, (2) 
where 
i
y is the actual daily tourist flow, 
i
y ˆ
 
is the 
predicted daily tourist flow, and N is the total number of 
test samples. 
After computing the RMSE of the three prediction 
algorithms based on the sample data with and without the 
Baidu index feature, a t-test was employed to assess the 
influence of the Baidu index feature on algorithm 
performance and compare the performance of the three 
prediction algorithms. A t-test p value below 0.05 
indicates a significant difference, while a value below 0.01 
signifies an extremely significant difference. 
3.5 Experimental results 
Before utilizing the samples to train each machine 
learning algorithm, this paper analyzed the correlation 
between the selected feature indicators and daily tourist 
flow to ensure that these indicators can affect the flow. 
Table 4 shows that all p values were less than 0.01, 
signifying a significant correlation between the nine 
feature indicators and daily tourist flow. The result 
verified the feasibility of using these nine feature 
indicators for predicting daily flow. 
Table 4: Results of the correlation analysis between the 
sample feature indicators and daily tourist flow. 
Feature 
indicator 
Correlation 
coefficient 
t p 
Date type 
t
s 1.11 1.36 0.001 
Week type 
t
w 
0.98 1.11 0.002 
Number of 
holiday days 
t
h 
1.35 2.14 0.001 
Average 
tourist flow 
s
y 
0.87 0.58 0.001 
Month 
t
m 0.52 0.87 0.002 
Antecedent 
daily tourist 
flow 
k t
y
−
 
1.37 1.14 0.001 
Weather 
condition 
t
a 
1.25 0.87 0.001 
The standard 
deviation of 
tourist flow 
s
v 
1.12 0.99 0.002 
Baidu index 
t
se 
2.47 1.58 0.000 
 
Due to space limitations, only the prediction results of 
the three prediction algorithms for April 2021 are 
presented in Figure 2. The daily tourist flow at Longkou 
Nanshan Scenic Spot fluctuated between 10,000 person-
time and 30,000 person-time during April 2021. Upon 
comparing the prediction outcomes of the three 
algorithms, it was found that the results of the SVM and 
traditional BPNN algorithms exhibited a noticeable 
deviation, with the SVM algorithm displaying a relatively 
more significant deviation. In contrast, the results obtained 
by the LSTM algorithm generally aligned with the actual 
daily flow, with only occasional deviations on some days. 
Moreover, these deviations were smaller compared to the 
other two algorithms. 

Predicting Daily Tourist Flow in Scenic Areas Using LSTM…                                     Informatica 48 (2024) 33–40   37 
 
 
Figure 2: Results of the three forecasting algorithms for 
daily tourist flow in April 2021. 
Comparing the mean square error of the three 
algorithms with or without the Baidu index feature and 
assessing the differences' significance, the results are 
summarized in Table 5, with specific values omitted here. 
Table 5 shows that when predicting daily tourist flow 
using the same algorithm, the RMSE and MAPE in the 
case without the Baidu index feature was significantly 
higher than that with the Baidu index feature, especially 
the LSTM algorithm. When comparing different 
prediction algorithms under the same feature indicators, it 
is observed that the RMSE and MAPE of the three 
prediction algorithms differed significantly. Regardless of 
the presence of the Baidu index feature, the SVM 
algorithm exhibited the highest RMSE and MAPE, 
followed by the traditional BPNN algorithm, while the 
LSTM algorithm had the lowest. 
Table 5: The RMSE and MAPE of the three prediction 
algorithms with and without the Baidu index feature and 
the significant degree of differences. 
Algorithm Without the 
Baidu index 
feature 
With the Baidu 
index feature 
Indicator RMSE MAP
E 
RMS
E 
MAPE 
SVM 35837.
65 
22.89 2963
8.56* 
20.13* 
BPNN 26789.
66 
17.54 2154
7.74* 
15.22* 
LSTM 15787.
58 
12.11 1258
9.87* 
10.14* 
P value 
between SVM 
and BPNN 
0.043 0.041 0.048 0.045 
P value 
between SVM 
and LSTM 
0.001 0.000 0.001 0.000 
P value 
between 
BPNN and 
LSTM 
0.001 0.001 0.001 0.001 
Note: * indicates a significant difference within the same 
kind of algorithm with or without the Baidu index feature, 
i.e., the p value is less than 0.05. 
The robustness test results of the LSTM algorithm are 
shown in Table 6. It can be seen that there was no 
significant difference in the errors of daily tourist flow 
forecasts for different scenic areas using the LSTM 
algorithm. In other words, the LSTM algorithm algorithm 
can be applied to predict daily tourist flow for other scenic 
areas. 
Table 6: The robustness test results of the LSTM 
algorithm. 
 RMSE MAPE 
Longkou Nanshan Scenic Spot 12589.87 10.14 
Longkou People’s Park 12578.57 10.21 
Huangshui River Wetland 
Park  
12547.98 10.24 
The p value between the 
comparison of the Longkou 
Nanshan Scenic Spot and 
Longkou People’s Park 
0.125 0.214 
The p value between the 
Longkou Nanshan Scenic Spot 
and Huangshui River Wetland 
Park  
0.215 0.187 
The p value between the 
Longkou People’s Park amd 
Huangshui River Wetland 
Park 
0.236 0.158 
4 Discussion 
 
This article used the LSTM algorithm, a deep learning 
algorithm, to forecast the daily tourist flow. In order to 
further enhance the accuracy of the prediction, the Baidu 
index feature was also introduced. Then, simulation 
experiments were conducted on the proposed algorithm 
and compared with the SVM and BPNN algorithms. 
Firstly, a correlation analysis was performed on nine 
indicators related to daily tourist flow, and the analysis 
results showed that all the indicators were significantly 
related to daily tourist flow. The date type includes 
working days, weekends, and holidays. Typically, tourists 
have more time to travel on weekends or holidays, 
resulting in an increase in visitor flow at scenic spots. 
Therefore, there is a correlation between date type and 
visitor flow. The weekday type refers to a specific day of 
the week, similar to the date type. Generally, tourists are 
more likely to travel on Saturdays and Sundays, so there 
is a correlation between weekday type and tourist flow. 
The holiday duration refers to the length of vacation time. 
Usually, longer vacations tend to promote tourist travel 
and increase tourist flow at scenic spots; hence there is a 
correlation between holiday duration and visitor flow. The 
average tourist flow represents the average number of 
visitors at different time periods and directly reflects 
changes in visitor flow at scenic spots; therefore it has a 
correlation with tourist flow. “Month” refers to the month 
to which a date belongs, and for scenic spots, there are 

38   Informatica 48 (2024) 33–40                                                                                                   F. Li 
peak seasons and off-peak seasons. The number of tourists 
is higher during the peak season, so there is a correlation 
between the month and visitor flow. The antecedent daily 
tourist flow refers to the number of visitors in a certain 
period before the current day, which also correlates with 
tourist flow prediction. The weather condition refers to the 
weather on that day; better weather conditions can 
promote tourist activities. The standard deviation of 
tourist flow reflects the fluctuation in tourist numbers 
during a specific time period. Baidu index, an indicator 
reflecting online user behavior data, was used in this 
article to reflect tourists’ searches and evaluations for 
scenic spots. The more searches and evaluations there are, 
the more attractive the scenic spot is to tourists. 
In comparison with the other algorithms, the results also 
showed that the introduction of Baidu index as a feature 
improved the accuracy of the prediction algorithm. 
Additionally, the LSTM algorithm used in this study 
demonstrated higher predictive accuracy compared to the 
other two algorithms. The reason for this is that Baidu 
Index reflects internet users’ behavior data in information 
retrieval, which can reflect tourists’ interest in scenic spots 
on the Internet and further supplement the features of 
tourist behavior data, thereby enhancing the accuracy of 
predictions. Furthermore, compared to the SVM and 
BPNN algorithms, the LSTM algorithm not only uses the 
activation function in hidden layers to fit non-linear 
patterns but also utilizes the gate mechanism to 
incorporate historical information, resulting in higher 
predictive accuracy. 
The novelty of this article lies in the use of LSTM 
algorithm to forecast daily tourist flow with time series 
characteristics. To enhance the precision of prediction, 
Baidu index was introduced as an effective reference for 
accurately predicting daily tourist flow in scenic areas. 
The limitation of this article lies in the insufficient span of 
the algorithm's prediction for daily tourist flow, or rather, 
its instability in predicting large-span daily tourist flow. 
Although robustness tests have been conducted, the 
dataset lacks diversity in terms of types. Therefore, future 
research directions include further improving the accuracy 
and generalization ability of the algorithm. 
5 Conclusion 
This paper provides a concise overview of relevant 
indicators for predicting daily tourist flow at tourist 
attractions. The Baidu index data from search engines was 
introduced into these indicators. The LSTM algorithm was 
employed for prediction. Moreover, simulation 
experiments were conducted to compare the LSTM 
algorithm with the traditional BPNN and SVM algorithms 
using the Longkou Nanshan Scenic Spot, spanning 
January 2017 to December 2022. Key findings are as 
follows. Nine feature indicators, including date type, week 
type, month, number of holiday days, average tourist flow, 
antecedent daily tourist flow, weather condition, the 
standard deviation of daily tourist flow, and Baidu index, 
exhibited significant correlations with daily tourist flow. 
The SVM and traditional BPNN algorithms displayed 
noticeable deviations in daily tourist flow prediction, with 
the former showing a more significant deviation; in 
contrast, the prediction results of the LSTM algorithm 
generally aligned with the actual daily tourist flow, 
exhibiting only occasional deviations. When the identical 
prediction algorithm was used to predict daily tourist flow, 
the RMSE and MAPE were significantly higher for 
feature indicators without the Baidu index feature than 
those with the Baidu index feature. Additionally, among 
the three prediction algorithms under the same feature 
indicators, the SVM algorithm had the highest RMSE and 
MAPE, followed by the traditional BPNN algorithm, and 
the LSTM algorithm had the lowest. 
The article used the LSTM algorithm to forecast the 
daily tourist flow with time series characteristics. In order 
to enhance the accuracy of prediction, it also introduced 
the Baidu index feature, which provides an effective 
reference for accurately predicting daily tourist flow in 
scenic areas. 
6 References 
[1] Zhang Z, Wang C, Gao Y, Chen Y, Chen J (2020). 
Passenger Flow Forecast of Rail Station Based on 
Multi-Source Data and Long Short Term Memory 
Network. IEEE Access, 8, pp. 28475-28483. 
https://doi.org/10.1109/ACCESS.2020.2971771 
[2] Guo Z, Zhao X, Chen Y, Wu W, Yang J (2019). 
Short-term passenger flow forecast of urban rail 
transit based on GPR and KRR. IET Intelligent 
Transport Systems, 13(9), pp. 1374-1382. 
https://doi.org/10.1049/iet-its.2018.5530 
[3] Liu W, Tan Q, Wu W (2020). Forecast and Early 
Warning of Regional Bus Passenger Flow Based on 
Machine Learning. Mathematical Problems in 
Engineering, 2020(1), pp. 1-11. 
https://doi.org/10.1155/2020/6625435 
[4] Guo Z, Wu HY, Yang CH (2015). Study on Method 
of BRT Transportation Station Passenger Flow 
Forecast with the TOD Mode. Applied Mechanics & 
Materials, 744-746, pp. 2059-2062. 
https://doi.org/10.4028/www.scientific.net/AMM.7
44-746.2059 
[5] Li Y, Ma C (2023). Short-Time Bus Route Passenger 
Flow Prediction Based on a Secondary 
Decomposition Integration Method. Journal of 
Transportation Engineering, Part A. Systems, 149(2), 
pp. 4022132.1-4022132.10. 
[6] Marie-Sainte SL, Saba T, Alotaibi S (2019). Air 
Passenger Demand Forecasting Using Particle 
Swarm Optimization and Firefly Algorithm. Journal 
of Computational and Theoretical Nanoscience, 
16(9), pp. 3735-3743. 
https://doi.org/10.1166/jctn.2019.8242 
[7] Lu T, Yao E, Liu S, Zhou W (2020). Short-time 
Forecast of Entrance and Exit Passenger Flow for 
New Line of Urban Rail Transit During Growth 
Period. Tiedao Xuebao/Journal of the China Railway 
Society, 42(5), pp. 19-28. https://doi.org/10.1049/iet-
its.2018.553010.3969/j.issn.1001-8360.2020.05.003 
[8] Yao E, Zhou W, Zhang Y (2018). Real-Time 
Forecast of Entrance and Exit Passenger Flow for 

Predicting Daily Tourist Flow in Scenic Areas Using LSTM…                                     Informatica 48 (2024) 33–40   39 
Newly Opened Station of Urban Rail Transit at 
Initial Stage. Zhongguo Tiedao Kexue/China 
Railway Science, 39(2), pp. 119-127. 
https://doi.org/10.3969/j.issn.1001-4632.2018.02.15 
[9] Wang Y, Ma J, Zhang J (2020). Metro Passenger 
Flow Forecast with a Novel Markov-Grey Model. 
Periodica Polytechnica Transportation Engineering, 
48(1), pp. 70-75. https://doi.org/10.3311/PPtr.11131 
[10] Wen K, Zhao G, He B, Ma J, Zhang H (2022). A 
decomposition-based forecasting method with 
transfer learning for railway short-term passenger 
flow in holidays. Expert Systems with Applications, 
189, pp. 1-16. 
https://doi.org/10.1016/j.eswa.2021.116102 
[11] Ni M, He Q, Gao J (2017). Forecasting the Subway 
Passenger Flow Under Event Occurrences With 
Social Media. IEEE Transactions on Intelligent 
Transportation Systems, 18(6), pp. 1623-1632. 
https://doi.org/10.1109/TITS.2016.2611644 
[12] Zhang J, Shen D, Tu L, Zhang F, Xu C, Wang Y, 
Tian C, Li X, Huang B, Li Z (2017). A Real-Time 
Passenger Flow Estimation and Prediction Method 
for Urban Bus Transit Systems. IEEE Transactions 
on Intelligent Transportation Systems, 18(11), pp. 
3168-3178. 
https://doi.org/10.1109/TITS.2017.2686877 
[13] He Y, Li L, Zhu X, Tsui KL (2022). Multi-Graph 
Convolutional-Recurrent Neural Network (MGC-
RNN) for Short-Term Forecasting of Transit 
Passenger Flow. IEEE Transactions on Intelligent 
Transportation Systems, 23(10), pp. 18155-18174. 
[14] Sajanraj TD, Mulerikkal J, Raghavendra S, Vinith R, 
Fabera V (2021). Passenger flow prediction from 
AFC data using station memorizing LSTM for metro 
rail systems. Czech Technical University in Prague - 
Central Library, 31(3), pp. 173-189. 
https://doi.org/10.14311/NNW.2021.31.009 
[15] Sha S, Li J, Zhang K, Yang Z, Wei Z, Li X, Zhu X 
(2020). RNN-based subway passenger flow rolling 
prediction. IEEE Access, 8, pp. 15232-15240. 
https://doi.org10.1109/ACCESS.2020.2964680 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

40   Informatica 48 (2024) 33–40                                                                                                   F. Li