Research on Security Incident Analysis: Scenarios and Challenges

Xuan Liu

1†

, Kehang Zhou

2†

, Qixuan Chen

3†

and Jianing Zheng

4†

Internet of Things Beijing University of Technology, Beijing, China

Software Engineering Chongqing University of Posts and Telecommunications, Chongqing, China

Guangdong Experimental High School Guangdong, China

Management Information System & Supply Chain Management University of Minnesota Twin Cities, Minnesota, U.S.A.

†

These authors contributed equally

Keywords: Event Detection, Risk Pre-warning, Systematic Review, Data Analysis.

Abstract: This paper systematically reviews the research of safety analysis, which includes the detection of adverse

events, the prediction of events, and the discovery of potential safety hazards. It mainly reviews the three

aspects that are suitable for people's lives, namely, facility safety, complete medical care, and the prediction

of the Internet of things. This study intends to analyze from the following dimensions: describing the current

safety accidents in people's lives and their potential hazards; how to predict the possibility of crisis;

summarizing the methods proposed in the literature. In addition, the limitations of the literature have been

summed up: the prediction method mentioned in this paper needs to be improved in a few aspects, which

means that might not meet the needs of some special cases; the prediction accuracy of logistics needs to be

improved; the error in medical treatment is difficult to avoid. This review has certain theoretical and practical

significance. To be theoretically speaking, relevant scholars are enabled to have a systematic understanding

of security analysis and help them to formulate the future research agenda. From reality, practitioners can

systematically understand the methods of safety detection and prediction, to formulate corresponding

preventive measures and emergency plans.

1 INTRODUCTION

In the history of human development, the surging

development of technology and science ensure many

aspects of people's life. People's travel and the

products that they use are all the outputs of the

development of them. Meanwhile, in our daily life,

the security of facilities also attracted people's

attention. During the process of using innovations,

people are more inclined to how to correctly use the

security facilities with higher efficiency at the same

time. There are a lot of existing literates on facilities

and safety prediction. Several articles are summarized

which are widely used in different fields so that

practitioners or scholars can understand the

significance of the safety field and the impact of it in

the future, as well as the benefits of ensuring the

safety of facilities to the development of human

society, however, there is no relevant literature

review. Therefore, the objective of this paper is to

systematically review the research of safety analysis,

in which safety analysis includes the detection of

adverse events, event prediction and the discovery of

safety hazards. This study intends to analyze from the

following dimensions: firstly, describing the current

safety accidents in people’s lives and their potential

hazards; second, how to predict the possibility of

crisis and the methods proposed in the cited literature;

third, the potential significance of this paper. This

review article mainly focuses on event security,

information prediction and the Internet of Things

based on historical data, including the fields of

medical and facility security, then the way of

predicting the possibility of the event that people

avoiding to let it happen based on the given

information. The purpose of this event is attributed to

the field of analysis, which includes the current

situation of the event and the prediction of the

probability of an accident in the future. This paper

introduces examples of event safety in different fields.

Besides, the technology that can be used in the

process of forecasting, the mathematical calculation

based on the historical model, and what needs to be

improved in selected articles have also been focused

on.

Liu, X., Zhou, K., Chen, Q. and Zheng, J.

Research on Security Incident Analysis: Scenarios and Challenges.

DOI: 10.5220/0011767100003607

In Proceedings of the 1st International Conference on Public Management, Digital Economy and Internet Technology (ICPDI 2022), pages 745-752

ISBN: 978-989-758-620-0

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

745

This review has several theoretical and practical

implications. From a theoretical point of view, it can

enable relevant scholars to have a systematic

understanding of safety analysis and help them to

shape future research agendas. From a practical point

of view, practitioners can systematically understand

the methods of safety detection and prediction so as

to formulate corresponding preventive measures and

emergency plans.

2 METHODS AND RESULTS

2.1 Search Strategy

The purpose of this paper is to help people predict

events more reasonably. So, to choose articles about

facilities and the Internet of things, most of the

previous articles are about medicine. Therefore, in

order to avoid paying too much attention to the

medical field, the articles at the bottom of the article

list are excluded. The main sources of articles are

Google, Google Scholar, and school libraries. After

searching for keywords like Event Prediction, Event

Security, Event Detection, and Facility Prediction, the

most relevant articles with a preference for the

number of references are collected. Eventually,

articles about medical, facilities and IoT were

obtained.

2.2 Security Analysis in Public Facility

Scenarios

To create a comfortable, reasonable, and safe living

environment for residents with an effective plan of

development of the country, decision-makers have to

work hard to come up with better solutions to achieve

this goal. This part briefly introduces three kinds of

daily life scenarios which need to be guaranteed

safety and the technology applied in this paper,

furthermore, a wireless sensor application technology

is also explained. Those methods are used to provide

reference ideas for decision-makers.

2.2.1 Event Detection in the Crowded Video

Most of the dangerous emergencies occur in the

crowd which prone to the crowded dangerous

situation in real life (Ke 2007). So to ensure people's

safety, this paper provides some methods to detect

dangerous events by using the video of the

technology. But this method also has some errors. For

example, first, the tracker is often affected by the

noise of the environment; second, the 2D modeling

has a similar disadvantage - it can't capture some

motions and objects; third, the model is built

manually and is derived from a single sample of

events, thus it is impossible to predict the change of

events, some problems are not involved in this model.

The experimental process of this method will be

described in the following paragraph.

This experiment takes the object being tracked as

a speck, based on observations detection of the

distance between the position of it and the spot.

Further, the event which has spatial interaction

activities are classified with training vertical and

horizontal vectors, processing the sequence in the

video. In this method, the volume of video space is a

3D object, so that different objects in the video can be

distinguished - the object in events produce different

shapes. The next step is shape matching which is used

to identify events to detect the distance. This

algorithm needs to set a minimum threshold, using

sliding window detection technology to keep the error

rate as less as possible.

Another way is to use unsupervised clustering

technology to extract the contour of spatial shape

model in the video, that is used to segment objects and

backgrounds, besides, the background might have

Optical Flow often makes noise which often causes

errors. The model build in this method is combined

with Shechtman and Irani's Flow Consistency method

(Ke 2007), which is a complementary feature, so it

can be calculated corresponding functions in the

complex environment with graph and ground

separation, to reduce the interference. To match and

recognize between objects and the model, the author

introduces a baseline matching algorithm of volume

shape matching model based on parts so that the

method organizes the part model into a tree structure

and models the relative position of each part as a

Gaussian distribution with diagonal covariance

matrix to improve the matching efficiency.

The event is detected in the shape model of time

and space in this method, demonstrating how to do

shape matching in a cluttered scene with dynamic

background, and it complements the previous

dynamic description to calculate in the complex

environment so that it can be extended to 3D for event

recognition. This technology realizes low precision

and recalls near all actions, and the behavior based on

the part shape and description flow is better than

previous studies, which shows the powerful

performance of the recorded video.

ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology

746

2.2.2 Smart City

With the development of the times, the impact of

computing public infrastructure on the smart city is

crucial, such as supermarkets, hospitals, office space,

to meet the different needs of people. These affect

people's way of life, at the same time, the vehicles

people drive will also affect the changes in the

facilities, for example, traffic flow at different times.

Using sensor technology to obtain the movement

mode in space can find the problems faced by the city

and provide a new perspective to overview the

modern city, so that helps the city planners make

more reasonable decisions. There are two challenges

at the beginning (Wang, et al. 2016): the first is how

to effectively calculate the influence degree when the

data is a large number of vehicles and facilities; the

second is how to predict the influence when the

current position of vehicles is known. Based on these

two aspects, the experiment uses the following

methods to achieve the goal.

This article (Wang, et al. 2016) using a network

(grid) index method is to map the position of a vehicle

to a specific unit. In this unit, the number of vehicles

will be added to the area which is the nearest facility

around it. That means although each vehicle has its

trace, when the tracks of the two cars are similar, the

destination that they are going to is homologous. So a

model based on the Markov Chain is used, which is

one of the various versions of it, that can be learned

through the historical trail. This model predicts the

movement of vehicles in the following days so that

designers can get the range of more, it is used to

change and update the planning of future facilities.

Index method of the grid. It is a kind of grid that

divides the geographical area into the same size and

provides an index for each grid. Trace can be replaced

by a grid, which is represented by an index in the grid.

In this model, the scope and coverage of different

facilities are different due to the size, so the number

of grids covered by it is also variance. The number of

taxis in the grid and then get the impact of this facility

on the city based on the qualities of each facility are

calculated. Although this method seems remarkably

simple, there may be facilities crossing range that

reflect the nearby grid. And the size of grid selection

also affects the distribution of future devices to a

certain extent. However, this method also has

advantages, it can be used to predict the impact of

facilities management, meanwhile can also avoid

many issues, such as traffic jams, furthermore, it can

solve data separately in predicting future location.

The previous Markov model uses the way of

transferring the state of each grid to the matrix, and

its working rate is trained by the number of historical

steps of movements which means this method is

changing the data from one unit to the corresponding

matrix. Besides, the number of transferring efficiency

is used for the transition probability between two

units. The basic Markov model is to predict unknown

traces. Another situation is that if the traces are

known, the Bayesian theorem can help to calculate

the probability of these objects will go to the

destination (Wang, et al. 2016).

This experiment provides how to predict the

influence degree of facilities in the smart city

according to the traffic flow and predict the future

movement through grid division and historical

training model. Although the prediction location may

introduce uncertainty and affect the prediction results,

as long as the distance between the prediction grid

and the actual distance is not far, and around the same

public facilities, the impact is small. The influence

and precision of the experiment are acceptable. So,

people can use the method mentioned in the article to

predict a public facility to meet people's needs for

urban life.

2.2.3 High Voltage Transmission System

People's daily life is inseparable from the voltage

system. To maintain continuous supply, High Voltage

Transmission Systems (HVTS) need to be maintained

reliably (Jaya 2019). Because the system failure will

lead to huge economic losses, but also endanger

people's lives and use. Therefore, in the process of

maintenance, the safety of a high voltage

transmission system is of great significance in the real

scene. This paper introduces the safety assessment

and control of the voltage system. According to the

failure rate of a historical model, the possibility of

future failure is predicted, besides, the accuracy of

this method is as high as 89.88%. Next, the

implementation of this method will be focused on.

This experiment (Jaya 2019) applies a kind of K-

neural network technology which is a simple and

efficient way to be used in object recognition. So, in

various searching for research, it is common to use

for prediction. One of the main methods is that K-and

can be used to classify and evaluate by decision trees,

discriminant analysis, and logistic regression. This

function offers active training and searching for the

best classification model which has almost 90%

accuracy, but the speed is at a medium level around

all the functions of finding. The method adapts

various metrics to determine the distance, according

to the given set 'X' of 'n' points, the KNN nearest

searching can help to find the point which is closest

Research on Security Incident Analysis: Scenarios and Challenges

747

to the target in X. This way is widely used in machine

learning. In check the safety of the facility, there will

be less time on finding the place that needs to be

repaired.

To predict events, the more mistake made, the less

efficiency people will have. Thus, the correct range

of calculating is important. One of the solutions is that

using the failure rate of the power system which can

be expressed by Gaussian function by using a fuzzy

security index binary- the least square method. And

this way is to find the state of keeping the facility run

normally.

𝜇







𝑥



=exp−(

(

𝑒



−𝑥





2𝜎





) (1)

In general, this method uses the Markov model

which is divided into 16 states for modeling and is

divided into five security levels. The accuracy of

prediction is determined by Safety Index. The

conclusion of the experiment is according to the

exception and the prediction is accurate. Through this

method, this method can be applied to more facility's

safety prediction to ensure the safety of the staff.

2.2.4 Wireless Technology

Wireless sensor technology is used for detection and

prediction. At the same time, this paper (Zhong, et al.

2015) demonstrates that sound is predictable, and

introduces a fuzzy logic system based on the sound

signal that can be a prediction. Also, a double sliding

window detection model is applied to reduce the error.

Wireless sensors can monitor machine facilities.

If the sound increases abnormally, it indicates that the

machine may appear some mistakes, this can be seen

as an event. So, machines can be check and predicted

by the wireless sensor with sound signals.

Variance time chart is widely used in verifying the

time series. The common character of the sound

signal is automorphism, which is the principle of the

following method adapted by the author. So, this way

can be applied to the prediction. A fuzzy logic system

is also a kind of time series prediction that can be used

to calculate intensive time. It can capture the

characteristics of the sound signal made by the

machine, which means that it can be predicted by

historical behavior and then check. For the potential

application of prediction of the sound signal, in this

paper, a new algorithm applying for reducing errors

of event detection - 'double sliding window event

detection' is proposed, this method is used to regulate

noise range in the uncontrollable environment and

determine a fixed threshold of different surroundings.

To alleviate the problem of a range selection.

The method (Zhong, et al. 2015) is superior to the

historical model. Through the signal-to-noise ratio

and self-discipline of sound, besides, the error is

reduced by using the time variance model - fuzzy

logic system and double sliding window detection.

The experiment proves that the wireless sensor can be

used to predict and detect the related security after

simulating the related experiment models.

2.3 Sequential Prediction on IoT

Scenarios

Nowadays, the Internet of Things (IoT) is becoming

more and more important in our life. The number of

devices connected to the IoT will reach 26 billion

devices by 2020, according to Gartner Group. The

Internet of things is the inevitable trend of this era. In

this part, the prediction of IoT will be discussed. Due

to the characteristics of the Internet of things in time

and space, the prediction of IoT will be necessary.

There are two kinds of prediction of IoT to discuss.

The first prediction used Bayesian Network and

association rules to predict associated events and the

second prediction used clustering and event detection

in social streams to do some predictions.

This part uses two algorithms to do some

sequential predictions on IoT. This sequential

prediction will be particularly useful in IoT. It can

predict delays in one transport phase provides

sufficient time for rescheduling activities in the next

and some goods that the user wants to buy can also be

predicted according to some of his shopping records

or logistics records and can also be predicted through

some purchase records of other users.

2.3.1 Bayesian Networks

Literature adopts a Bayesian network, which is used

to build a model with uncertainty (Karakostas 2016).

It is a directed acyclic graph, and each of its nodes

represents a discrete random variable. There is a

causal relationship between the variables represented

by its child node and those represented by its parent

node. The edge between two nodes represents the

probability dependence between events.

𝑒



->𝑒



: means the possibility of 𝑒



depends on e2, P

(𝑒



)! = 0.

𝑃

(

𝐴

𝐵

)

𝑃

(

𝐴

)

𝑃

(

𝐵

𝐴

)

𝑃

(

𝐵

)

(2)

In this case, it is assumed that event B is fixed,

that is,𝑃

(

𝐵

)

=1. The impact of the determined event

B on event A is needed to be considered.

𝑃

(

𝑒



𝑒



)

𝑝



(

𝑒



)

∗𝑝

(

𝑒



𝑒



)

𝑝

(

𝑒



)

(3)

ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology

748

Suppose that 𝑒



and 𝑒



are two potential causal

objects. This causality determines the subsequent

events triggered by the event generated by the object.

Statistical knowledge will be used to build a

conditional probability model and use the Bayesian

formula to give the above formula.

𝑃

(

𝑒



𝑒



)

= 𝑝



(

𝑒



)

∗𝑝

(

𝑒



𝑒



)

(4)

𝑝



(

𝑒



)

is the probability of observing E2

independently of other events, P ( 𝑒



- 𝑒



) is the

influence intensity of 𝑒



on 𝑒



probability. If E1 is

assumed to be 𝑃（𝑒



） = 1, the above formula is

obtained

Since some events can cause a chain reaction,

some events can be decomposed. For example, event

𝑒 can produce chain reactions 𝑒



, 𝑒



, 𝑒



, ..., 𝑒



. These

n intermediate events are linked to calculate the

probability of 𝑒



's dependence on event 𝑒 .

P ( 𝑒



| 𝑒



) is calculated the conditional

probability of 𝑃

(

𝑒



𝑒



)

2.3.2 Association Rules

Literature also adopts association rules (Rudin, et al.

2011), and the idea of the algorithm is as follows.

Algorithm 1: Subroutine GenRules.

- Input: (S, B, X ), that is, past orders S =

{zi}i=1,...,m, zi ⊆ X , current basket B ⊂ X , set of

items X

- Output: Set of all rules



𝑎𝑗→𝑏𝑗



𝑗 where 𝑏𝑗 is a

single item that is not in basket B, and where aj is

either a subset of items in basket B, or else it is the

empty set. Also, the left-hand side 𝑎𝑗 must be allowed

(meaning it is in A). That is, output rules {aj → bj}j

such that bj ∈ X \B and aj ⊆ B ⊂ X with aj ∈ A, or aj

= ∅.

Algorithm 2: Max Confidence, Min Support

Algorithm

- Input: (θ, X , S, B, GenRules, c), that is,

minimum threshold parameter θ, set of items X , past

orders S = {zi}i=1,...,m, zi ⊆ X , current basket B ⊂

X , GenRules generates candidate rules GenRules (S,

B, X ) = {aj → bj}j , number of recommendations c ≥

- Output: Recommendation List, which is a subset

of c items in X

- Flow of the algorithm：

a) Apply GenRules (S, B, X ) to get rules {aj →

bj}j where aj is in the basket B and bj is not.

b) Compute score for each rule aj → bj as ¯fS,θ

(aj , bj ) = fS,0 (aj , bj ) = # (aj∪bj ) #aj when support

#aj ≥ θ, and ¯fS,θ (aj , bj ) = 0 otherwise.

c) Reorder rules by decreasing score.

d) Find the top c rules with distinct right-hand

sides, and let Recommendation List be the right-hand

sides of these rules

Algorithm 3: Adjusted Confidence Algorithm.

- Input: (K, X , S, B, GenRules, c), that is,

parameter K, set of items X , past orders S =

{zi}i=1,...,m, zi ⊆ X , current basket B ⊂ X ,

GenRules generates candidate rules GenRules (S, B,

X ) = {aj →

bj}j , number of recommendations c ≥ 1

- Output: Recommendation List, which is a subset

of c items in X

- Flow of the algorithm：

a) Apply GenRules (S, B, X ) to get rules {aj →

bj}j where aj is in the basket B and bj is not.

b) Compute adjusted confidence of each rule aj →

bj as fS,K (aj , bj ) = # (aj∪bj ) #aj+K .

c) Reorder rules by decreasing adjusted

confidence.

d) Find the top c rules with disti

2.3.3 General Architecture of IoT Prediction

This diagram is the general architecture of IoT

prediction. Different events are generated by the IoT

cloud and stored through the capture interface

supporting the M2M protocol of the typical IoT. In

the storage warehouse, events are cleaned up and

classified, and then the category probability of events

is calculated according to the above model. Since new

events are added to the warehouse periodically, the

probability is recalculated periodically. The time

prediction model is connected with the decision

support system.

2.4 Clinical Predictive Models in

Medical Scenarios

As medical equipment plays a more and more

important role in modern hospitals, the safety

detection and development of medical equipment has

become an important link in hospital management.

Different medical devices play different roles

depending on the severity of the disease. Some of the

devices are just adjunct to treatment and health care,

while most of the devices are lifesaving. Take dialysis

equipment for example. The patient's life is supported

by it. More than 350,000 patients receive dialysis in

the United States, and the safety of their care is

ultimately the responsibility of the medical director of

the facility. Dialysis facilities are a complex

organization that involves multiple disciplinary

providers and uses advanced technologies to care for

patients with a wide range of serious diseases. As

organizations become more complex and the

Research on Security Incident Analysis: Scenarios and Challenges

749

likelihood of errors increases, potential risks must be

identified and prioritized. To investigate the attitudes

of dialysis patients and health care providers toward

safety through a randomized national review of

dialysis patient data. According to the findings, safety

risks mainly come from patient safety hazards (such

as communication or machine design), dialysis

equipment failures, and medication errors.

Improve and change health care safety through the

use of QAPI programs. A successful QAPI program

requires a reliable collection and rigorous analysis of

data. One widely accepted form is the plan-dot-

check-action cycle. This approach first "plans" the

required process improvements with the available

data and literature, and then implements those

improvements ("do"). Evaluate the results to

determine whether performance has improved (the

"check"). If the results improve, the results are shared

and the new process ("act") is implemented across the

organization, and the results are re-validated to ensure

that the improvements are effective and ongoing. If

the expected results are not met, alternative or

additional improvements must be planned and

implemented, and the results reanalyzed.

In order to provide effective outcome prediction

for patients and provide a good basis for treatment

decisions, clinical prediction plays a very important

role. Clinical predictive models play an important

role in medical decision-making, such as risk

stratification and treatment allocation. Their

effectiveness needs to be tested systematically in

large scale trials. They have developed a dynamic

clinical prediction model with discrete time to event

data with competitive risk, and they extended

standard objective Bayesian variable selection

methods to address discrete-time competitive risk

models and identified the most relevant predictors of

ventilator-associated pneumonia (VAP) caused by a

specific microbe, Pseudomonas aeruginosa (PA).

By studying the competitive events between

pneumoviruses to provide more effective pr-

0ediction and a good basis for new patients. Data on

new patients from different periods and patients from

different centers were obtained by recording 5123

days of intensive care units. This study uses the

Bayesian variable (discrete data processing) selection

method to deal with the discrete-time competition risk

model. The discrete-time event models are

collectively called logistic regression.

𝛿





1 𝑇



≤𝐶



0 𝑇



> 𝐶



𝜆



(

𝑡

𝑥



)

=Pr

(

𝑇



= 𝑡, 𝑅



= 𝑟

𝑇



≥𝑡, 𝑥



)



















∑





















, for 𝑟=1,…,𝑚, 𝑡= 1,2, …, (5)

Table 1: Summary findings of Renal Physicians Association health and safety surveys.

Safety Issues during the

Prior 3 Months

Professional Staff

Response (%)

Patient Response (%)

Patients worried or concerned

about safety/staff

Patients communicated

concerns sometimes or

always: 63

Sometimes or always worried:

Ease of communication

Easy to communicate with

patients: 94

Uncomfortable / somewhat

uncomfortable communicating

with staff: 18

BP or weight not recorded prior

to dialysis

Happened sometimes: 13 Happened sometimes: 10

Mistakes in membrane or bath

set up

Happened sometimes: 60 Happened sometimes: 6

Lapses in infection control

(hand hygiene)

Reported event occurred: 27 Reported event occurred: 11

Medication errors

Missed or incorrect dose

occurred sometimes: 23

Always discuss all

medications with staff: 23

Difficulty with access to needles

Rare or no difficulty

inserting: 66

Pain at access site during

treatment: 39

Prolonged access to bleeding Sometimes: 15 Sometimes: 23

Needle dislodgement prior to

the end of treatment

Sometimes occurred: 4 Reported event occurred: 5

Medical mistakes in prior 3

months

Reported no events

occurred: 70

Reported no events occurred:

Used with permission from the RPA Health and Safety Survey

ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology

750

However, once ventilator-assisted or extubated

patients died during the trial, they were no longer at

risk of VAP hypertension. In addition to

discrimination, another important aspect of prediction

model evaluation is calibration, which informs the

reliability of the forecast risk. Calibration

performance will be evaluated using a calibration

diagram (CP) and calibration slope (CS) and intercept

(CI).

log 

𝜆



(

𝑡

𝑥



)

𝜆



(

𝑡

𝑥



)

= 𝑎



+ 𝑏

,

𝑛







(6)

In addition to clinical medicine, more specifically,

it extends to cardiovascular disease. The researchers

investigated whether arterial pulse wave velocity

(aPWV) could improve the temporal prediction of

cardiovascular disease. They looked at 17, 635 people

with heart problems and collected individual data

from 16 studies. Proportional risk models and random

effects models assessed the mixed effects and

determined the study association between APWV and

cardiovascular disease outcomes. They looked at 17,

635 people with heart problems and collected

individual data from 16 studies. Proportional risk

models and random effects models assessed the

mixed effects and determined the study association

between APWV and cardiovascular disease

outcomes. The researchers conducted a systematic

review and used newly published and unpublished

cohort data, including APWV and cardiovascular

events, to perform an individual participant meta-

analysis. Their objective was to determine whether

APWV information could improve the prediction of

future cardiovascular events in unselected,

population-based individuals and patients with

significant disease; whether the risk prediction is

different among factor groups; whether it improves

the risk. The results of this experimental study suggest

that AWVP may be a useful biomarker to improve

cardiovascular risk prediction in moderate-risk

patients.

3 DISCUSSION

3.1 Main Findings

In general, the significance of security prediction for

human life and development from three dimensions

of security facilities, Internet of things and medicine

have been discussed. Some conclusions are found in

these three different dimensions.

First, in the security facilities area, some video

recognition technology and voice recognition

technology can be used to predict what may happen

in our daily life, and apply them to the smart city to

provide convenience for people's daily life. In the

power system, it is also possible to predict the

potential dangers in the high-voltage power network

through some technologies.

Second, in the IoT prediction area, some of the

collected detection and prediction technologies can be

applied to the prediction of the Internet of things and

use Bayesian theory and some association rules to

predict the sequential events and apply them to the

Internet of things. For example, the places prone to

logistics congestion can be predicted through the

logistics records of many users and try our best to

avoid these problems in the subsequent logistics

routes Lines.

Third, in the field of medicine, some prediction

can help doctors to make some reference to the

patient's condition, and also can predict some

potential threats of patients in some treatment process

to help doctors find and predict in time and reduce the

occurrence of the risk.

3.2 Limitations and Suggestions for

Future Work

On safety detection, event detection from videos

cannot be used for all kinds of detection, because the

model of it is from a single example, this could limit

its ability to summarize the changes of different

events (Ke 2007). This paper tries to use this method

in the smart city (Wang, et al. 2016), and this method

still needs to improve. Besides, the smart city has

another problem which is difficult to find the biggest

influential equipment when all the objects are moving

(Wang, et al. 2016). The errors exist, but they are

acceptable at present. Because of that, in the future,

there is still more work to do especially on modeling

which including auto-selecting events and

aggregating multiple training videos into a single

model so that this method can get better

generalization performance (Ke 2007). The security

prediction of a high voltage transmission system can

make some objective predictions for the safety of

power system facilities, but at present, some chemical

elements are still uncontrolled, it can still not predict

the influencing factors (Jaya 2019). Therefore, the

security problems of the power system from the

chemical level or using some chemical methods can

be predicted. The wireless acoustic sensors cannot

determine the range of relevant thresholds because the

noise level of the detected object is unknown, there is

a method called the double sliding window algorithm

that can fix this problem (Zhong, et al. 2015).

Research on Security Incident Analysis: Scenarios and Challenges

751

On the IoT prediction, the Bayes theorem can be

used to do some IoT predictions. In this paper,

research that used association rules is collected to

predict sequential events (Rudin, et al. 2011) and

researching that predicted events through social

stream (Aggarwal 2012). However, all the methods

can not 100% sure to predict an event. They are more

likely to be reference resources. The IoT needs

abundant events because it is dynamic (Karakostas

2016). Besides, some of the events which cannot be

observed, their associated events for observation and

prediction are needed to find (Rudin, et al. 2011). In

the future, when the IoT is more completed, the

prediction will be more accurate, but the ability to

grab keywords, as well as the accuracy of clustering

problems should be improved, to be able to improve

the accuracy of prediction problems in the IoT.

In the medical area, there will inevitably be errors

in the prediction. For example, there are

uncontrollable variables in the process in bio-

manufacturing facilities such as the environmental

factors (Yang, Farid and Nina 2014). The next

problem is the heart problem (Ben-Shlomo, et al.

2014), the age and other diseases can affect the

accuracy of the results. The medical facilities can be

improved in the future, and computer technology and

medical technology should be connected more in the

future and make full use of the advantages of

computer technology.

4 CONCLUSION

To sum up, this paper has analyzed security events in

different fields, including detection and prediction.

The paper introduces the relevant safety accidents in

different reasonable fields and the learning methods

with techniques used to detect the possibility of safety

accidents. It mainly summarizes the most general

safety accidents in public places, for example,

platforms and medical fields, and how to forecast

them to avoid errors, meanwhile improving the

reliability. The significance of this study focuses on

analyzing security and providing efficient solutions.

Future research can be strengthened in data selection,

universality and adapting the latest prediction

technology. Besides, this study might be further

applied to more fields to ensure human security, such

as information, health and other aspects.

REFERENCES

Aggarwal, Charu C., and Karthik Subbian. "Event detection

in social streams." Proceedings of the 2012 SIAM

international conference on data mining. Society for

Industrial and Applied Mathematics, 2012.

Ben-Shlomo, Yoav, et al. "Aortic pulse wave velocity

improves cardiovascular event prediction: an individual

participant meta-analysis of prospective observational

data from 17,635 subjects." Journal of the American

College of Cardiology 63.7 (2014): 636-646.

Jaya, C. K., R. Sunitha, and Abraham T. Mathew. "Security

prediction of high voltage transmission system (HVTS)

based on k-nearest neighbor (k-NN) classifier

technique." Journal of Intelligent & Fuzzy Systems 36.6

(2019): 5773-5782.

Ke, Yan, Rahul Sukthankar, and Martial Hebert. "Event

detection in crowded videos." 2007 IEEE 11th

International Conference on Computer Vision. IEEE,

2007.

Karakostas, Bill. "Event prediction in an IoT environment

using naïve Bayesian models." Procedia computer

science 83 (2016): 11-17.

Rudin, Cynthia, et al. "Sequential event prediction with

association rules." Proceedings of the 24th annual

conference on learning theory. JMLR Workshop and

Conference Proceedings, 2011.

Wang, Hongtao, et al. “Influential Spatial Facility

Prediction over Large Scale Cyber-Physical Vehicles in

Smart City.” EURASIP Journal on Wireless

Communications and Networking, vol. 2016, no. 1,

2016, doi:10.1186/s13638-016-0606-4.

Yang, Yang, Suzanne S. Farid, and Nina F. Thornhill. "Data

mining for rapid prediction of facility fit and

debottlenecking of biomanufacturing facilities."

Journal of biotechnology 179 (2014): 17-25.

Zhong, Zhen, et al. "Security assurance in wireless acoustic

sensors via event forecasting and detection." Security

and Communication Networks 8.16 (2015): 2671-2677.

ICPDI 2022 - International Conference on Public Management, Digital Economy and Internet Technology

752