257
h ps:// esea ch endsjou nal.com
Online a : h ps:// esea ch endsjou nal.com ISSN No: 2584-282X
Indexed Jou nal Pee Re iewed Jou nal
INTERNATIONAL JOURNAL OF TRENDS IN EMERGING RESEARCH AND DEVELOPMENT
Volume 2; Issue 6; 2024; Page No. 257-261
Recei ed: 08-08-2024
Accep ed: 19-10-2024
In usion de ec ion sys ems o SQL da abases using machine lea ning
1Chadchanka Ama na h Shi anand, 2D . Bal ee Singh, 3D . Yashpal Singh, 4D . Rohi a
Yamagan i and 5D . Swa i Dey
1Resea ch Schola , P.K. Uni e si y, Shi pu i, Madhya P adesh, India
2-5P o esso , P.K. Uni e si y, Shi pu i, Madhya P adesh, India
DOI: h ps://doi.o g/10.5281/zenodo.17132174
Co esponding Au ho : Chadchanka Ama na h Shi anand
Abs ac
SQL Injec ion is s ill among he wo s secu i y laws, exposing use s' p i a e da a and causing inancial losses. The mos cu en OWASP
Top 10 s udy anks injec ion a acks as he op ulne abili y, and he equency o hese a acks is on he ise. By de ini ion, In usion
De ec ion Sys em (IDS) ules ha ely on s a ic signa u es a e a common componen o adi ional de ensi e sys ems. These ules a e g ea
o p e en ing known a acks bu a en' e ec i e agains ze o-day h ea s. Many ecen s udies ha e used machine lea ning me hods, which
may iden i y p e iously unseen h ea s bu can be pe o mance-hea y depending on he me hodology. To add insul o inju y, some in usion
de ec ion sys ems scou da abase se e logs o in o ma ion, while o he s ocus on ga he ing a ic en e ing he web app ac oss he ne wo k
o he web app hos . A web applica ion hos , a MySQL da abase se e , and a Da iphy appliance node placed be ween he wo a e he wo
sou ces o a ic ha will be collec ed in his p ojec . Th ough ou examina ion o hese wo da ase s as well as an addi ional da ase ha is
co ela ed wi h hem, we ha e p o en ha he accu acy achie ed wi h he co ela ed da ase using algo i hms like decision ees and ule-
based app oaches is compa able o ha o a neu al ne wo k algo i hm, bu wi h subs an ially be e pe o mance.
Keywo ds: Indus y, adi ional, machine lea ning, pe o mance, companies
In oduc ion
Web assaul s like SQL Injec ion ha e been a ound o a
long ime, bu hey s ill endange indi iduals' p i a e
in o ma ion as well as he secu i y o companies and
go e nmen s su e inancial losses as a esul . Tha is
especially he case when conside ing new a ack ec o s a e
always appea ing and old ones a e cons an ly e ol ing. Web
a ack mi iga ion ecei es subs an ial unding om indus y
and secu i y businesses; ye , many exis ing mi iga ion
solu ions ha e limi s ha a e being ac i ely sough o be
o e come by cu en esea ch.
S a ic analysis o incoming online a ic, o en called one
ypical app oach o adi ional online a ack mi iga ion is
signa u e de ec ion. C ea ing a signa u e ha is speci ic o
cybe a acks is he essence o his s a egy; when a i ewall
o o he secu i y appliance de ec s his signa u e, po en ially
obs uc he suspicious a ic. Al hough his me hod is
quick and may be used in eal- ime o p o ec ne wo k
asse s, i does ha e one d awback: i can only iden i y
known assaul s.
One o he way o p o ec agains SQL injec ion a acks on
he web is o examine he syn ax o incoming SQL que ies;
an a ack o his kind is iden i ied when a que y is ound o
be e oneous. Un o una ely, his echnique equi es
ex ensi e unde s anding o he p og am and he amewo k
o "o dina y" ques ions; ye , i manages o do a an as ic job
o de ec ing no el h ea s using inco ec que ies and has
good de ec ion esul s o e all.
One app oach o de ec ing SQL injec ions ha is now being
in es iga ion use echniques de i ed om machine lea ning.
The employmen o SVMs, decision ees, neu al ne wo ks,
and ule-based lea ning echniques is common in his ield
o s udy. The abili y o iden i y no el assaul s is a majo
s eng h o hese me hods. One possible downside o hese
s a egies is ha , depending on he algo i hm u ilized,
p ocessing ime migh po en ially ise.
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
258
h ps:// esea ch endsjou nal.com
In usion De ec ion Sys ems
One o he main unc ions o an IDS is o keep an eye ou
o any suspicious o illegal beha io on a ne wo k o inside
i s sys ems. Usually, In he e en ha a secu i y inciden
occu s, a SIEM sys em will ei he ale he adminis a o o
compile all o he pe inen da a in o a single loca ion o an
incu sion o b each. By combining da a om many sou ces
and using ale il e ing algo i hms, a SIEM sys em can
dis inguish be ween legal and haza dous ac i i y.
In usion de ec ion sys ems' many sphe es o in luence a y
om indi idual PCs o ex ensi e ne wo ks. These days,
NIDSs and HIDSs a e he wo main ca ego ies used o
desc ibe hese kinds o sys ems. An HIDS would be a
sys em ha keeps abs on c i ical OS iles, whe eas an NIDS
would be one ha examines all incoming ne wo k a ic.
Ano he way o ca ego ize IDS is by de ec ion me hod. Two
o he mos popula a ia ions a e signa u e-based de ec ion
and anomaly-based de ec ion a e wo me hods o de ec ing
ha m ul pa e ns and abno mal a ic, espec i ely. The
o me employs machine lea ning o ind signa u es, while
he la e analyses his o ical da a o iden i y de ia ions om
a model o "good" a ic. One mo e popula a ia ion is
epu a ion-based de ec ion, which in ol es iden i ying
possible h ea s based on hei epu a ion a ings. The
capaci y o eac o de ec ed in usions is a ea u e o se e al
IDS solu ions. In usion p e en ion sys ems (IPS) a e o en
used o desc ibe sys ems ha can espond o po en ial
in usions. The use o specialized ools may also make
in usion de ec ion sys ems mo e use ul; o example, a
honeypo can be used o de ec and iden i y malicious
a ic.
Machine Lea ning
Machine lea ning asks
The e a e p ima ily wo s ages o any c ea ing an algo i hm
o machine lea ning. Di e en models ha e di e en ways
o being ained. Common amewo ks used o his pu pose
and o designing machine lea ning models include sciki -
lea n, Tenso low, PyTo ch, Ma lab, and Weka. When i
comes o op imizing he me hod o he numbe o
pa ame e s accessible, he amewo k is king.
Among he a ious applica ions o machine lea ning
models, in usion de ec ion inds special a en ion in h ee:
classi ica ion, eg ession, and econs uc ion. En ies a e
so ed in o many classi ica ions by classi ica ion, which
migh include "no mal" o "a ack" o e en dis inc amilies
o assaul s. In o de o ind con inuous alues, such is he
possibili y ha a gi en inpu cons i u es an assaul ,
eg ession (also e e ed o as "p edic ion") is used. Las ly,
only a ce ain class o neu al ne wo ks can do
econs uc ion. In o de o ge ep esen a ion lea ning allows
he ne wo k o acqui e he cha ac e is ics, his job a emp s
o decomp ess and ecomp ess he inpu da a in an e o o
ebuild i .
Machine lea ning algo i hms may be ained in one o wo
ways: supe ised o unsupe ised. Neu al ne wo ks a e one
example o a model ha may be augh in ei he di ec ion.
The majo i y o models a e ained ia supe ised lea ning,
whe e he da ase includes inpu s and he espec i e co ec
esul s. Assigning hese ou pu s o hei espec i e inpu s is
a ma hema ical unc ion, which he algo i hm lea ns o
ep esen . Classi ica ion and eg ession a e wo classic
supe ised aining p oblems. Con a ily, indings om he
aining da ase a e no used in unsupe ised aining.
Familia izing onesel wi h in iguing da a s uc u es is i s
goal. An example o an ac i i y ha does no need
supe ision is econs uc ion.
Machine lea ning models need es ing once aining is
comple e in o de o e alua e hei e icacy. Da a ha wasn'
in he aining se can' be used o his assessmen . I he
model had seen his da a be o e—o he igh ou comes in
he case o supe ised lea ning— hen he assessmen would
be skewed. You can also compa e al e na i e pa ame e s'
alues ( o example, a lea ning a e o neu one coun ) using
a alida ion se . Finding he alue ha pe o med he
subsequen op pe o me on he alida ion se s ep a e
aining. We nex pu he whole ne wo k h ough i s paces
on he es da a. Addi ionally, esh da a is equi ed o a
alida ion se ; ypically, a subse se aside speci ically o
his pu pose.
Table 1: Supe ised lea ning s unsupe ised lea ning
Supe ised lea ning
Unsupe ised lea ning
P ocess
Inpu and ou pu da a a e gi en.
Only inpu da a is gi en.
Inpu da a
The machine is ained using labeled da a.
The machine is no gi en unlabeled da a.
Algo i hms used
SVM, NN, Random Fo es , Linea and Logis ics
eg ession, Classi ica ion ees
Di e en ca ego iez: K-means, Clus e
algo i hms, Hie a chical clus e ing, and so on.
Compu a ional complexi y
Simple.
Complex.
USA o da a
Uses aining da a and ela e inpu and ou pu esul s.
Does no use ou pu da a.
Accu acy o esul s
Accu a e and us wo hy.
Less accu a e and us wo hy.
Real ime lea ning
Lea ning is o line.
Real- ime.
Numbe o classes
Known.
Unknown.
Main d awbacks
Big da a is a challenge.
No p ecise in o ma ion in ega ds o da a
so ing, and he ou pu is no known.
Li e a u e Re iew
Deepa Manikandan e al. (2024) [1] Secu i y ac oss all
domains, including da abases, ne wo ks, and he cloud, has
g own in impo ance in eal- ime dis ibu ed sys ems since
he ad en o cybe space. When aced wi h e ol ing h ea s,
p esen -day in usion de ec ion sys ems (IDS) ind i
di icul o s ay in ol ed. The p oposed model DR-DBMS
(dimensionali y educ ion in da abase managemen sys ems)
combines supe ised machine lea ning me hods,
dimensionali y educ ion app oaches, and sophis ica ed ule-
based classi ie s o enhance he accu acy o in usion
de ec ion o di e en kinds o assaul s. Simula ion indings
show ha he DR-DBMS sys em e ec i ely used me hods
o dimensionali y educ ion and ai s selec ion o
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
259
h ps:// esea ch endsjou nal.com
de e mining in usion assaul in 0.07 seconds wi h a lesse
numbe o cha ac e is ics.
Roland Plaka (2021) [2] In usion de ec ion sys ems (IDS)
a e designed o keep an eye ou o any b eaches in a
ne wo k's con iden iali y, in eg i y, o a ailabili y ha may
ha e occu ed as a esul o malicious o unapp o ed
ac i i y. Machine lea ning me hods o de ec ion and
classi ica ion, a a ie y o in usion de ec ion sys ems (IDS),
and anomaly de ec ion app oaches we e ho oughly
e iewed in his hesis. We p o ide an a chi ec u e o
in usion de ec ion sys ems (IDS) ha combines adi ional
me hods wi h machine lea ning echniques. The cu en
de ec ion me hods may be enhanced o be e a ack
de ec ion and ca ego iza ion by including sui able machine
lea ning echniques. Addi ionally, we ha e made an e o o
e alua e and execu e a ba e y o i ual es s on each
machine lea ning algo i hm o compa e hei pe o mance.
Fo gene al in usion de ec ion sys ems used in indus ial
con ol applica ions, ou me hod o e s indica ions o
choosing machine lea ning echniques.
Yasmeen S. Almu ai i e al. (2022) [3] Sa egua ding
sophis ica ed communica ion ne wo ks equi es in usion
de ec ion sys ems (IDS). Ce ain pa e ns, signa u es, and
ule b eaches we e he p ima y a ge s o hese sys ems'
design. Po en ial new me hods o ne wo k in usion
de ec ion ha e eme ged Using Deep Lea ning and Machine
Lea ning me hods in he las many yea s. Techniques like
his may ell he di e ence be ween egula pa e ns and
hose ha a en' . Suppo Vec o Machine, J48, Random
Fo es , and Naï e By es we e among he machine lea ning
echniques used o assess he Ne wo k In usion De ec ion
Sys ems (NIDS) in his a icle. The me hods used bina y
and mul i-class ca ego iza ion on he NSL-KDD benchma k
da a se . Ex ensi e discussion is p o ided abou he
ou comes o using such s a egies, which exceeded ou
ea lie e o s.
Ami Singh e al. (2024) [4] In usion de ec ion sys ems
(IDS) a e acing new hu dles due o enc yp ed da a, new
p o ocol a ie y, and an inc ease in he amoun o c iminal
ac ions globally. In usion de ec ion sys ems ha ely on
signa u es a e now a a s age whe e inadequa e in his case.
Se e al academics ha e pu o h IDSs ha use machine
lea ning o de ec in usions p e iously unseen ha m ul
ac ions by analysing pa e ns o ac i i y. In usion de ec ion
sys ems ha use on machine lea ning ha e many ad an ages
han SIDS ha ely on signa u es a e mo e e ec i e in
spo ing no el o ms o ne wo k-based malwa e. This s udy
e alua ed ne wo k in usion de ec ion sys ems u ilising he
IDS da ase , which con ains he mos ecen common
a acks, and wo da a esampling echniques in addi ion o
10 machine lea ning classi ie s. The op h ee IDS models-
XGBoos , KNneighbo s, and AdaBoos -pe o m be e han
bina y-class classi ica ion wi h 99.49%, 99.14%, and
98.75% accu acy, espec i ely. XGBoos , KNneighbo s,
and GaussianNB a e he mos accu a e in mul i-class
classi ica ion wi h 99.30%, 98.88%, and 96.66% accu acy,
espec i ely.
Tahsinu Rahman (2022) [5] The need o secu i y measu es
o p e en b eaches is g owing in di ec co ela ion wi h he
p oli e a ion o in e ne -connec ed de ices. The pu pose o
an In usion De ec ion Sys em (IDS) is o iden i y and block
po en ially ha m ul ne wo k a ic. The need o anomaly-
based in usion de ec ion sys ems ha use machine lea ning
o iden i y mo e ecen assaul s has a isen because new
h ea s may easily bypass s anda d signa u e-based IDS.
This hesis will ocus on a speci ic heme: anomaly-based
in usion de ec ion sys ems ha use deep lea ning. A
compa ison is made be ween ad e sa ial machine lea ning
me hods and Gene a i e Ad e sa ial Ne wo ks (GAN), wi h
con en ional deep lea ning echniques. We use s a is ical
me ics on wo sepa a e da ase s o assess he s a egies. As
pa o he assessmen p ocess, malicious samples a e aken
in o accoun alongside benign and known a ack samples.
The las s ep is o compa e he op imal me hod o o he
open-sou ce anomaly-based in usion de ec ion sys ems.
Ou pe o ming all o he echniques was a me hod ha used
a GAN o gene a e ad e sa ial samples. Fu he mo e, when
aced wi h malicious da a poin s, he me hod ma ches he
pe o mance o cu en anomaly-based IDS. A e e iewing
he li e a u e, we ha e come o he conclusion ha in usion
de ec ion sys ems based on GANs may be enhanced o
be e wi hs and bo h new and malicious assaul s.
Resea ch Me hodology
To de e mine whe he incoming communica ion is benign o
malicious, ou p oposed me hod in his s udy employs
machine lea ning me hods. A bespoke business cha web
app unning on a dis an MySQL se e o ms he backbone
o he sys em. The e a e wo loca ions whe e da a is
eco ded: i s , in he HTTP a ic ha lows be ween he
se e s ha gene a e a ic and hose ha hos he web
applica ions. Secondly, in he da a ans e be ween he
webapp se e and he o line da abase se e using
MySQL.
Fig 1: Sys em P ocess
SCADA Da ase
A e in oducing he issue and ML me hods u ilized his
sec ion o he chap e aims o p o ide a gene al ou line o
he i s da ase . This da a collec ion o igina es om a CI
wa e sys em ha is con olled by SCADA.
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
260
h ps:// esea ch endsjou nal.com
Fig 2: SCADA Sys em A chi ec u e
Expe imen and Resul s
Table 1 summa izes he esul s o ou esea ch wi h se e al
machine lea ning echniques. The ime i akes o cons uc
he machine lea ning models is called Model Time, he ime
i akes o classi y he es ing da ase using 5- old c oss-
alida ion is called Tes ing Time, and Accu acy is he
classi ica ion accu acy ha each me hod achie es. Figu e 1
displays he classi ica ion accu acy, whe eas Table 2
displays he F-sco es.
Table 2: Resul s Wi h 20000 Reco ds
Da ase
Algo i hm
Accu acy
Model
Tes ing
Time
Time
Webapp
JRip
94.740%
3m30.85s
2.70s
J48
95.630%
1m42.65s
2.55s
RF
96.525%
5m30.20s
30.60s
SVM
94.025%
2m35.80s
1m10.45s
ANN
96.715%
46m03.55s
3.35s
Da iphy
JRip
95.980%
6m10.90s
3.35s
J48
96.995%
1m59.30s
2.45s
RF
97.210%
6m10.20s
33.50s
SVM
95.190%
2m11.50s
1m3.45s
ANN
97.285%
41m23.00s
2.95s
Co ela ed
JRip
97.150%
11m50.80s
2.10s
J48
97.295%
2m01.80s
2.05s
RF
98.055%
4m22.55s
35.45s
SVM
95.715%
3m30.85s
1m5.90s
ANN
97.615%
47m25.25s
3.80s
Scada
To demons a e he p ecision o anomaly de ec ion, his
sec ion de ails and assesses h ee sepa a e s udies. The
pu pose o he a ious es s is o o e a ying deg ees o
de ail abou he p esence o an abno mali y. F om jus
no ing ha an abno mali y has occu ed o pinpoin ing he
o ending pa and he unusual ci cums ance, he e is a wide
ange o possible ou comes.
Table 3: SCADA Resul s: Expe imen 1 - Anomaly De ec ion (5-
old c oss- alida ion)
Classi ica ion (Is Anomaly)
Recall
P ecision
F1-Scp e
LR
Benign
7.15%
90.34%
13.22%
Anomaly Weigh ed
A e age
99.89%
87.7%
93.4%
87.73%
88.05%
88.05%
NB
Benign Anomaly
99.95%
16.74%
28.67%
24.99%
99.97%
39.98%
Weigh ed A e age
34.82%
89.06%
89.06%
k-NN
Benign
74.01%
79.7%
76.74%
Anomaly
97.15%
96.12%
96.63%
Weigh ed A e age
94.12%
93.97%
93.97%
SVM
Benign
7.15%
92.24%
13.23%
Anomaly
99.91%
87.70%
93.41%
Weigh ed A e age
87.75%
88.30%
88.30%
Ke nel SVM
Benign
39.53%
98.52%
56.40%
Anomaly
99.91%
91.63%
95.59%
Weigh ed A e age
91.99%
92.545£
92.54%
DT
Benign
74.01%
74.72%
74.35%
Anomaly
96.22%
96.09%
96.15%
Weigh ed A e age
93.30%
93.28%
93.28%
RF
Benign
75.66%
75.99%
75.78%
Anomaly
96.38%
96.33%
96.36%
Weigh ed A e age
93.67%
93.67%
93.67%
MQTT: We use he six ML algo i hms-LR, Gaussian NB,
k-NN, SVM, DT, and RF-discussed p e iously o e alua e
a ious ML app oaches on he MQTT-IoT-IDS2020 da ase .
The ollowing cha ac e is ics a e disabled o p e en any
iden i ying da a om being impac ed: p o ocol, MQTT
lags, sou ce and des ina ion IP add esses.
In e na ional Jou nal o T ends in Eme ging Resea ch and De elopmen h ps:// esea ch endsjou nal.com
261
h ps:// esea ch endsjou nal.com
Table 4: MQTT-IoT-IDS2020: O e all De ec ion Accu acy
Packe
Fea u es Unidi ec ional
Bidi ec ional
LR
k-NN
DT
78.87%
98.23%
99.44%
69.13%
99.68%
99.9%
88.55%
99.96%
99.95%
RF SVM (RBF
Ke nel)
65.39%
99.98%
99.97%
77A%
97.96%
96.61%
NB SVM (Linea
Ke nel)
81.15%
78%
9755%
66.69%
82.6%
98.5%
Fig 3: MQTT-IoT-IDS2020: O e all De ec ion Accu acy T end
using Di e en ML Techniques
Conclusion
Secu i y o sensi i e in o ma ion, including inancial and
heal h eco ds, emains a op conce n due o SQL injec ion
a acks and o he web-based assaul s. This p oblem is
becoming mo e p essing as mo e and mo e social ac i i ies
ely on he in e ne . We ha e demons a ed ha he
algo i hms we ha e es ed, including ule-based and
decision ee algo i hms, can classi y es ing da a
signi ican ly as e and wi h accu acy compa able o ha o
Neu al Ne wo ks, and we ha e also sugges ed a mul i-
sou ce da a analysis sys em o imp o e he accu acy o SQL
injec ion a ack de ec ion. Adap ing his sys em o de ec
o he ypes o web-based a acks is on he lis o u u e
wo ks. O he hings on he lis include ga he ing mo e da a,
like a ic going ou bound om he web app o he b owse ,
collec ing la ge da ase s o see i ha helps pe o mance,
and analyzing addi ional machine lea ning echniques o
accu acy and pe o mance.
In usion De ec ion Sys ems (IDS) a e p og ams ha scan
e e y incoming and ou going da a packe s o signs o
malicious ac i i y. In he las en yea s, IDS ha e been buil
using a a ie y o ML app oaches. The apid eme gence o
new cybe h ea s has led o he widesp ead usage o ML
app oaches. This hesis del es in o he use o ML
app oaches o cons uc IDS wi h speci ic pu poses. In
addi ion, his hesis explo es he possibili y o enhancing he
e iciency and e icacy o nex -gene a ion IDS by using
inno a i e DL app oaches ha ha e been success ully used
in o he a eas o s udy.
Re e ence
1. Manikandan D, e al. Machine lea ning app oach o
in usion de ec ion sys em using dimensionali y
educ ion. In e na ional Jou nal o In o ma ion
Technology. 2022;34(1):1-7.
2. Plaka R, e al. In usion de ec ion using machine
lea ning o indus ial con ol sys ems. Jou nal o
Indus ial In o ma ion In eg a ion. 2021;24:100223.
3. Almu ai i Y, Alhazmi B, Munshi A. Ne wo k in usion
de ec ion using machine lea ning echniques. Ad ances
in Science and Technology Resea ch Jou nal.
2022;16(3):193-206. doi:10.12913/22998624/149934.
4. Singh A, P akash J, Kuma G, Jain P, Amba i L.
In usion de ec ion sys em. Jou nal o Da abase
Managemen . 2024;35(2). doi:10.4018/JDM.338276.
5. Rahman T. In usion de ec ion sys em based on deep
lea ning. 2022 Jul 29. [Thesis/Repo ].
6. Sama L. Ne wo k in usion de ec ion using deep
lea ning. [Mas e ’s hesis]. Melbou ne: Vic o ia
Uni e si y; c2022.
7. Kül ü E. Ne wo k in usion de ec ion wi h a deep
lea ning app oach. 2022 Feb. 99 p. [Thesis/Repo ].
8. Wo ku G/Michael K. A p edic i e model o ne wo k
in usion de ec ion sys ems using machine lea ning
app oach. 2023 Jan. [Repo ].
9. Raka M, e al. In usion de ec ion using machine
lea ning and log analysis. In e na ional Jou nal o
Ad anced Resea ch in Compu e and Communica ion
Enginee ing. 2019;8(4):220-225.
10. Vanin P, Newe T, Dhi ani LL, O’Connell E, O’Shea D,
Lee B, Rao M. A s udy o ne wo k in usion de ec ion
sys ems using a i icial in elligence/machine lea ning.
Applied Sciences. 2022;12(22):11752.
doi:10.3390/app122211752.
C ea i e Commons (CC) License
This a icle is an open access a icle dis ibu ed unde
he e ms and condi ions o he C ea i e Commons
A ibu ion (CC BY 4.0) license. This license pe mi s
un es ic ed use, dis ibu ion, and ep oduc ion in any
medium, p o ided he o iginal au ho and sou ce a e
c edi ed.
