SMS SPAM DETECTION

Shivanand Aurasang Abhishek K Vasanthkumar

Dept. of ECE Dept. of ECE Dept. of ECE
KLE Technological University KLE Technological University KLE Technological University
Dr. MSSCET, Dr. MSSCET, Dr. MSSCET,
Belagavi Campus Belagavi Campus Belagavi Campus
[email protected] [email protected] [email protected]

Ritin Shimpi Shweta Madiwalar Neelam Somannavar

Dept. of ECE Dept. of ECE Dept. of ECE
KLE Technological University KLE Technological University KLE Technological University
Dr. MSSCET, Dr. MSSCET, Dr. MSSCET,
Belagavi Campus Belagavi Campus Belagavi Campus
[email protected] [email protected] [email protected]

Abstract—Machine learning based SMS spam identification it addressed by keyword matching. Mes- are continually being
has become a critical weapon protecting user security. giving tweaked by spammers, words that get sages to bypass rigid
up privacy and removing uninvited and potentially dangerous keyword filters and exploit their. lack of ability to recognize
communications. To detect spam messages in SMS This study in-
vestigates how to apply classification to datasets. techniques. They hairy or evolving spam patterns. As a result, the traditional
perform preprocessing, feature extractions as well as On model approaches give rise to high false-. where legitimate messages
training, the system hopes to distinguish accurately at the impasse are misclassified as positive rates. false positives, where actual
between spam and legitimate messages. This Research assesses spam doesn’t get caught and spammish, where we are right
by performance parameters as such as accuracy, precision, and about most of our total spam but aren’t exact undetected.
recall.concrete classifiers like Naive Bayes, or Support Vector
Machine I bring that experience, as well as my experience with The need for more sophisticated solutions that can be adapted
(SVM) and Random Forest in practice. By reducing exposure it for other situations. However, these limitations motivate ma-
to spam communications and improve automated text classifiers. chine learning (ML) to become a transformative technology.
This approach is to improve user experience. Index Terms—SMS Unlike static keyword- ML based methods learn dynamically
spam detection, Machine learning. Text classification fication from They can adapt to the ever changing landscape with
algorithms
Index Terms—SMS spam detection, Machine learning, Classi-
data. of spam messaging. Its random methods like Random
fication algorithms, Text classification will help you achieve this. There are three classifiers; Forest,
Support Vector Machine (SVM), and Naive Bayes. veloping
robust model using machine learning is possible. due to being
I. I NTRODUCTION
capable of analyzing and classifying text based features with.
It still is important to communicate through In today’s fast high precision.
paced digital age where there’s no alternative to convenience In today’s fast paced digital age, dealing in fast money online
for both personal and professional interactions and speed. It is convenient for both personal and professional interactions,
can facilitate quick updates, as well as coordination with col- and for speed. For example, it can facilitate quick updates,
SMS bridges gaps in leagagues, or even service notifications. or coordinate with col-league partners. SMS bridges gaps,
efficiently and effectively. But, the increase in It has also intro- or receiving service notifications in leagues. efficiently and
duced huge issues of unsolicited SMS spam of and security as effectively. Still, the rise in prevalence However, challenges
well as user experience. They are often promotional messages of unsolicited SMS spam of tremendous proportions have
(or, rather, SMS spams) not only disrupts, but advertisements, been introduced. It affects both user experience and security.
or even fraudulent schemes It communicates but also leads In the form of promotional messages, SMS spam is quite
to possible security threats. Phishing attacks, malware and commonplace. in my opinion, not only interrupts but also
breaches make up these categories. It can cause financial advertisements, or even fraudulent schemes. But at the same
loss or data theft; of privacy. The current rise in SMS spam time this enables communication but exposes users to potential
underscores the need for efficient detection mechanism. The security. threats. They include phishing attacks, malware, and
problem isn’t merely It’s a growing security risk, about incon- breeches. which can result in loss of money or theft of data.
venience, it’s a security risk that needs to occur. proactive, and The urgency of this escalation in SMS spam. This necessitates
reliable. Simple traditional spam filtering methods are often for effective detection mechanisms. The problem isn’t merely
used but those methods. However, these challenges are not it’s a growing security risk that demands about inconvenience.
reliable and proactive solutions. Typically relying on simple D. How Feature Selection Affects SMS Spam Detection
filters, traditional spam filtering methods like these These Given this, the further information for SMS spam detection
challenges are beyond the bounds of keyword matching. Mes- systems will be effective is the necessity of feature engi-
spammers always adapt with new techniques. sages that skip neering. The raw texts are often pre processed for analysis
past strict keyword filters and take advantage of it has inability by special techniques such as tokenization, stemming, and n-
to understand nuanced or evolving spam patterns. Thus, these gram. The paper shows that by providing the appropriate fea-
traditional approaches lead to a high false- legitimate messages tures in the model, feature selection increases the classification
are misclassified as positive rates. false positives where the models’ success rate by a considerable margin.An issue that
spam seems legitimate, and false negatives, where actual spam comes with it, however, is over reliance on certain qualities
slips through. undetected. The need for more is highlighted that might not be easily adaptable to other data sets. This
by this dual inefficiency. solutions that are sophisticated and underscores the importance of collective detailed assessment
adaptable. in order to identify the best features that can be identified for
II. LITRATURE REVIEW the validation of their impacts towards model.
A. Machine Learning Methods for SMS Spam Detection
SMS Spam Detection using Machine Learning methods The
use of which SMS spam detection has extensively investigated
and conventional machine learning methods such as Naive III. METHODOLOGY
Bayes.There are Support vector machine (SVM) and Decision
Trees. Because Naive Bayes is probabilistic, and well handles
text data. often it is selection due to its ease of use and efficacy
in text categorisation. Decision Trees offer interpretability
Support Vector Machines have done so through their hierar-
chical structure.Generating has shown good performance by
(SVM) These are hyperplanes that distinguish spam from
authentic messages.However, as spammers continue to find
new ways to spam, these means may struggle to adapt to spams
dynamic nature.
B. Using Deep Learning to Identify Spam in SMS
In This research paper focused on which they can spot
intricate patterns in deep learning techniques in particular,
Concurrent Neural and evolutional Neural Networks (CNNs)
In the field of SMS, network (RNNs) have become popular.
spam identification. RNNs can process sequential data well,
and are designed to understand C meant for the context of
messages, but CNNs work well With convolutional layers, we
pass local information. Studies have deep learned to deal with
spam detection accuracy better shown.invariably these learn-
ing models outperform conventional techniques. Nevertheless,
there are still issues such as the high processing power, and, Fig. 1. BLOCK DIAGRAM
most importantly, a requirement for large labelled datasets.
Then we have the difficulties in training and interpreting the A. Data Collection and Preprocessing
model.
Dataset: For the purpose of this study, we use the SMS
C. Combining different Models to detect Better SMS Spam Spam Collection dataset that features a rich number of SMS
Detection texts classified as ’spam’ and ’ham’ (which is not spam). This
SMS Spam Detection through Combing the Models These dataset is very important in training our spam detection model
models which blend machine learning and deep are hybrid. since it has a well balanced class distribution.
Unfortunately, learning techniques have become a viable tactic Preprocessing:First, to navigate past lexicon noise, we per-
to increase. How accurate is SMS spam identification. Re- form preprocessing where structural words such as punctuation
searchers have better performance indicators, such as precision marks, special characters and numbers are purge from the text
and recall they fuse the advantages of several algorithms data leaving only significant words. We then break the whole
including Deep learning techniques Random Forest. Hybrid communications into words and to make all the content lower
models Deep learning techniques’ deeper contextual awareness case as it is easier to work with.Next, we use feature extration
can be used, while using machine learning’s fast pattern using the Term Frequency-Inverse Document Frequency (TF-
recognition skills can take advantage of this. IDF) method to convert text into numbers representations
convenient for the machine learning techniques. we split the 2) Precision (P): The percentage of recovered messages
set further and mix it so it would create a random distribution that are pertinent to the user is known as precision.
of spam and ham messages, thus ensuring that any likelihood
of bias during It is especially different from other phases of
TP
the training and testing phase. Precision =
TP + FP
B. Augmentation 3) Recall (R): The percentage of successfully retrieved
messages that are pertinent to the query is known as recall.
To enhance the resilience of the developed SMS spam
detection model and for fixing the class imbalance in case
present in the dataset we use data augmentation strategies TP
Recall =
specifically applied to textual data. While standard picture data TP + FN
augmentation techniques such as rotatioN and zooming is not 4) F-Measure: It functions as the precision and recall
possible, then we can mimic augmentation which are synonym harmonic mean.
replacement, random insertion, word deletion. For example we
can replace particular words with a message with synonyms
in order to have variations in the same message. P×R
F-measure = 2 ×
P+R
C. Model Architecture 5) Matthews Correlation Coefficient (MCC): Even when
1. Data Collection: Obtain a labelled SMS dataset classified the sizes of the classes differ, MCC is used to assess the
as ”spam” or ”ham” from public or synthetic sources. quality of classification for two classes. It has a range of -1
2. Input the Message Dataset. Load the dataset into the system, to +1, with +1 denoting the best performance.
ensuring that it contains SMS texts and their related labels.
3. Data Pre-Processing: Remove special characters, tokenise,
lowercase, and eliminate stopwords to improve data quality. (TP × TN) - (FN × FP)
MCC = p
4. Text Classification: Convert text into numerical features (TP + FP)(TP + FN)(TN + FP)(TN + FN)
using vectorisation methods like Bag of Words or TF-IDF,
6) Spam Caught (SC): It is the ratio of detected spam to
then select the appropriate Naive Bayes variant.
all spam in the dataset.
5. Dataset Training: Using the labelled dataset, train the Naive
Bayes model to determine the probability of classifying mes-
sages as spam or ham.6. Test a New Dataset: Use a separate False Negative Data
SC =
test dataset to evaluate model performance by comparing Total Number of Spams
predicted labels to actual results.
7) Blocked Hams (BH): It is the ratio of the number of
hams that are detected as spam to the total number of hams
D. C. Evaluation Metrics
in the dataset.
The metrics assess the system’s ability to detect spam and
identify misclassifications. Some of the evaluation metrics are:
1) True Positive (TP): Positive events that are appropriately False Positive Data
BH =
classified are reported with a True Positive number. Total Number of Hams
2) False Positive (FP) refers to erroneously classified positive
instances, whereas False Negative. IV. RESULTS
3) (FN) refers to incorrectly classed negative examples.
4) True Negative (TN): This value indicates that negative
events were appropriately classified. The basic metrics listed
above were used to calculate various measures. We evaluated
spam detection system performance using metrics such as
accuracy (ACC), spam catch rate (SC), blocked hams (BH),
and Matthews.
1) Accuracy (ACC): Accuracy is defined as the proportion
of correctly categorized classes (True Positive and True
Negative) over the total number of classifications, as shown
by the formula below.

TN + TP
Accuracy = Fig. 2. Represents Accuracy comparision
TN + TP + FN + FP
Fig. 3. Represents Precision comparision

Fig. 6. Represents Confusion Matrix

TABLE I
C OMPARISON OF VARIOUS C LASSIFIERS WITH M ULTIPLE M ETRICS

Metric Naive Bayes Random Forest SVM AdaBoost

ACC% 97.94 98.12 97.94 96.86

BH% 94.41 93.99 94.15 92.23

Fig. 4. Represents Recall comparision SC% 99.37 99.79 99.48 98.74

F-Measure 0.926 0.931 0.926 0.887

MCC 0.914 0.922 0.915 0.870

Precision 0.960 0.986 0.966 0.920

Recall 0.894 0.882 0.888 0.857

Time(secs) 0.027 4.444 9.228 5.883

Fig. 5. Represents F-Measure comparision

AUC 0.984 0.991 0.990 0.968

TPR 0.894 0.882 0.888 0.857

 [11] Aliza, H.Y., Nagary, K.A., Ahmed, E., Puspita, K.M., Rimi, K.A.,
Khater, A. and Faisal, F., 2022, March. A comparative analysis of SMS
V. CONCLUSION spam detection employing machine learning methods. In 2022 6th Inter-
national Conference on Computing Methodologies and Communication
The era of knowledge and technology is upon us. People (ICCMC) (pp. 916-922). IEEE.
[12] Sharma, N., 2022, June. A methodological study of SMS spam classi-
are now abandoning the old forms of communication. Spam fication using machine learning algorithms. In 2022 2nd International
is a severe and vexing issue that affects these information Conference on Intelligent Technologies (CONIT) (pp. 1-5). IEEE.
and communication channels. The dataset was classified [13] Bosaeed, S., Katib, I. and Mehmood, R., 2020, April. A fog-augmented
machine learning based SMS spam detection and classification system.
using a variety of methods: Indian material was included In 2020 fifth international conference on fog and mobile edge computing
in our modified SMS spam collection dataset. According to (FMEC) (pp. 325-330). IEEE.
the findings, Multinomial Naive Bayes and Support Vector [14] Huang, T., 2019, October. A CNN model for SMS spam detection.
In 2019 4th International Conference on Mechanical, Control and
Machine are two of the top classifiers for detecting SMS Computer Engineering (ICMCCE) (pp. 851-85110). IEEE.
spam. . [15] Mathew, K. and Issac, B., 2011, December. Intelligent spam classifi-
cation for mobile text message. In Proceedings of 2011 International
Conference on Computer Science and Network Technology (Vol. 1,
pp. 101-105). IEEE.
VI. FUTURE WORK
Although it took a long time to complete, the classifier SVM Guidelines for writing and formatting conference papers are
with a linear kernel achieved the highest accuracy. Conversely, included in IEEE conference templates. Before submitting
MNB with Laplace smoothing had accuracy that was quite your conference paper to the conference, please make sure
similar to SVM, but it took a lot less time than SVM. The that all template text has been eliminated. Your paper might
greatest outcomes.and can Build throgh Block chain not be published if you don’t remove the template text from
. it.

R EFERENCES
[1] Srinivasarao, Ulligaddala, and Aakanksha Sharaff. ”Machine intelligence
based hybrid classifier for spam detection and sentiment analysis of SMS
messages.” Multimedia Tools and Applications 82.20 (2023): 31069-
31099
[2] Gadde, Sridevi, A. Lakshmanarao, and S. Satyanarayana. ”SMS spam
detection using machine learning and deep learning techniques.” 2021
7th international conference on advanced computing and communication
systems (ICACCS). Vol. 1. IEEE, 2021.
[3] Roy, Pradeep Kumar, Jyoti Prakash Singh, and Snehasish Banerjee.
”Deep learning to filter SMS Spam.” Future Generation Computer
Systems 102 (2020): 524-533.
[4] Uysal, Alper Kursat, et al. ”The impact of feature extraction and
selection on SMS spam filtering.” Elektronika ir Elektrotechnika 19.5
(2013): 67-72.
[5] Salman, M., Ikram, M. and Kaafar, M.A., 2024. Investigating Evasive
Techniques in SMS Spam Filtering: A Comparative Analysis of Machine
Learning Models. IEEE Access..
[6] Siagian, W., Setiadi, M.R. and Prasetyo, S.Y., 2023, August. Improving
SMS Spam Detection Through Machine Learning: An Investigation
of Feature Extraction and Model Selection Techniques. In 2023 In-
ternational Conference on Information Management and Technology
(ICIMTech) (pp. 288-293). IEEE.
[7] Manju, G., Mondal, S.J. and Sthapak, H., 2023, December. SMS Spam
Detection and Filtering of Transliterated Messages. In 2023 Intelligent
Computing and Control for Engineering and Business Systems (IC-
CEBS) (pp. 1-9). IEEE.
[8] Wijaya, E., Noveliora, G., Utami, K.D. and Nabiilah, G.Z., 2023,
August. Spam Detection in Short Message Service (SMS) Using Naı̈ve
Bayes, SVM, LSTM, and CNN. In 2023 10th International Conference
on Information Technology, Computer, and Electrical Engineering (IC-
ITACEE) (pp. 431-436). IEEE.
[9] Deepa, N., 2023, February. Accurate SMS Spam Detection Using
Support Vector Machine In Comparison With Linear Regression. In 2023
Fifth International Conference on Electrical, Computer and Communi-
cation Technologies (ICECCT) (pp. 1-4). IEEE.
[10] de Luna, R.G., Magnaye, V.C., Reaño, R.A.L., Enriquez, K.L., Astorga,
D., Celestial, T., Española, A.M., Lanting, B.A., Mugar, D., Ramos,
M. and Redondo, J., 2023, October. A Machine Learning Approach
for Efficient Spam Detection in Short Messaging System (SMS). In
TENCON 2023-2023 IEEE Region 10 Conference (TENCON) (pp. 53-
58). IEEE.