Detecting Malicious Bots in Social Media Accounts Using Machine Learning Technology.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
p-ISSN: 2395-0072
Volume: 10 Issue: 05 | May 2023 www.irjet.net
Detecting Malicious Bots in Social Media Accounts Using Machine
Learning Technology.
Megha Choudhar1, Pallavi Garje2, Deeksha Shinde3, Rutika Valanj4
1Mr. Dattatray Modani Professor, Dept. of Computer Engineering, P.E.S. Modern College of Engineering, Pune,
Maharashtra,India
1234UG Students, Dept. of Computer Engineering, P.E.S. Modern College of Engineering, Pune, Maharashtra, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In this age of technology boom, the application
of artificial intelligence and machine learning in everyday
life is increasing rapidly. Malicious bot activity is a
significant threat to the security and integrity of online
services and systems. Machine learning technology provides
a powerful tool for detecting and preventing such activity by
identifying patterns and characteristics of bot behavior. This
approach involves collecting large data sets of both benign
and malicious bot activities, extracting relevant features,
selecting appropriate classification algorithms, and
continuously monitoring the model in real time. Unique
concepts such as feature importance, correlation analysis,
domain knowledge, recursive feature elimination, wrapper
methods, and principal component analysis can be used for
feature selection. The choice of feature selection and
classification algorithm will depend on the specific use case,
the characteristics of the data, and the desired accuracy and
efficiency of the model. Using machine learning technology
to detect malicious bots provides a robust and effective
solution to protect online services and systems against this
growing threat.
Key Words: Social Media, Support Vector Machine
(SVM), Pattern recognition, Anomaly Detection,
Decision Tree.
1.INTRODUCTION
TThe internet has transform the way we live, work,
and communicate. However, with these benefits
come several challenges, one of which is the growing
threat of malicious bots. These automated software
programs can infiltrate websites, social media
platforms, and online services, causing damage to
businesses and individuals alike. In response,
researchers and security professionals have
developed a range of detection methods, including
traditional rule-based systems and more advanced
machine learning-based approaches. Machine-
learning technology has emerged as a powerful tool
for detecting and preventing malicious bot activity.
By leveraging algorithms and statistical models,
machine learning can analyze large datasets of bot
behavior, identify patterns and features, and classify
an activity as benign or malicious. This approach
offers several advantages over traditional methods,
including greater accuracy, efficiency, and scalability.
Furthermore, machine learning-based systems can
adapt and learn over time, improving their
performance and effectiveness. Despite the promise of
machine learning-based solutions, there are several
challenges and considerations that researchers and
practitioners must address. These include feature
selection, data preprocessing, algorithm selection, and
real-time monitoring. Moreover, the ethical and legal
implications of deploying machine learning for bot
detection must be carefully considered, including
issues related to bias, privacy, and accountability. In
this paper, we review the current state of the art in
malicious bot detection using machine learning
technology. We explore the various methods,
algorithms, and techniques employed in this field, and
discuss the benefits and limitations of each. We also
highlight some of the key challenges and
considerations that must be addressed to develop
effective and ethical machine learning-based solutions
for bot detection. Overall, we argue that machine
learning has the potential to significantly improve the
security and integrity of online services and systems,
and we call for continued research and innovation in
this important area.
2. LITERATURE SURVEY
1. Detecting Malicious Twitter Bots Using
Behavioral Modeling and Machine Learning
Techniques by Liu et al. (2019) - This paper
proposes a method for detecting malicious
Twitter bots using a combination of behavioral
modeling and ML techniques. The authors
collect data on a set of Twitter accounts and
use features such as posting behavior,
sentiment analysis, and network analysis to
train a set of classifiers, which are then used to
detect malicious bots with high accuracy.
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 173

p-ISSN: 2395-0072
2. Bot or Not? A Systematic Evaluation of Bot
Detection Methods in Twitter by Cresci et al.
(2017) - This paper presents a systematic
evaluation of different bot detection methods
for Twitter, including MLbased approaches.
The authors compare the performance of
different classifiers using a large dataset of
Twitter accounts and show that ML-based
approaches can achieve high accuracy in
detecting bots.
3. Botnet Detection Using Machine Learning
Techniques: A Survey by Alam et al. (2020) -
This paper provides a survey of different ML
techniques that have been used for botnet
detection, including both network-based and
host-based approaches. The authors review
the strengths and limitations of different ML
algorithms and highlight some of the
challenges in detecting botnets using ML
techniques.
4. Bot Detection in Online Social Networks: A
Survey by Almeida et al. (2019) - This paper
provides a comprehensive survey of different
methods for detecting bots in online social
networks, including MLbased approaches.
The authors review the different types of
features that can be used for bot detection,
such as network structure, content, and
behavior, and compare the performance of
different classifiers.
5. Machine Learning for Social Network
Analysis: A Survey by Shukla et al. (2020) -
This paper provides a survey of different ML
techniques that have been used for social
network analysis, including the detection of
bots. The authors review the different types
of features that can be used for social
network analysis, such as graph properties,
node attributes, and community structure,
and highlight some of the challenges in
applying ML techniques to social network
analysis.
3. PROPOSED SYSTEM
3.1 Problem Solving: Random Forest
Our proposed system detects malicious bots
on Twitter using machine learning
technology. It uses a Random Forest
algorithm for classification and involves data
collection, feature extraction and selection,
and performance evaluation. Random Forest
is effective for detecting bots due to its
ability to handle high-dimensional data and
non-linear relationships.
Formula: Random Forest algorithm can be
represented by the following formula:
y=f1(x)+f2(x)+...+fn(x).
In this formula, y represents the predicted
output of our Random Forest model, which is
whether a given Twitter account is legitimate
or malicious. x represents the input feature
vector, which consists of several selected
features that describe the characteristics and
behavior of the Twitter account. The decision
trees, f1, f2,..., fn are created using subsets of
the input features and data points. Each
decision tree is essentially a sequence of if-
else statements that recursively split the data
based on the selected features. The final
prediction is made by combining the
predictions of all decision trees.
For example, one possible decision tree in our
model might be:
 If the account age is less than 30 days
and the follower count is less than
100, the account is classified as
malicious.
 Otherwise, if the sentiment score of
the account's tweets is negative and
the retweet ratio is greater than 0.5,
the account is classified as malicious.
 Otherwise, the account is classified as
legitimate. By combining the
predictions of multiple decision trees
in this way, our Random Forest model
can achieve high accuracy and reduce
the risk of overfitting the training
data.
In the context of Twitter, some of the input
features that we might use in our model could
include:
 Account age: the number of days
since the account was created
 Tweet frequency: the number of
tweets that the account posts per day
 Sentiment score: the overall positive
or negative sentiment of the account's
tweets
 Bot network size: the number of other
accounts that are linked to the same
bot network
 Retweet ratio: the ratio of retweets to
original tweets for the account

p-ISSN: 2395-0072
By analyzing these features and training our
Random Forest model on a dataset of labeled
Twitter accounts, we can build an effective
system for detecting malicious bots on
Twitter. In summary, our proposed system
for detecting malicious bots on Twitter using
machine learning technology and the
Random Forest algorithm offers a promising
solution for improving the security and
integrity of social media platforms. By
leveraging advanced techniques such as
feature extraction and selection, we believe
that this system has the potential to
significantly reduce the impact of malicious
bot activity on Twitter and other social
media platforms
Fig -1: Flowchart for Random Forest Classification
3.2 Problem Solving: Decision tree
Our proposed system utilizes a combination of
feature engineering and the Decision Tree
Algorithm to effectively detect malicious bots on
Twitter. We start by collecting a large and
diverse dataset of Twitter accounts, including
both legitimate and malicious accounts, and
extract a range of features that are indicative of
bot-like behavior, such as activity level,
engagement rate, content similarity, and
metadata inconsistency.
Table -1: Feature selection in a Twitter dataset
To optimize the performance of the Decision
Tree Algorithm, we first preprocess the data by
removing any noise and outliers and then
perform feature scaling and normalization to
ensure that each feature has the same weight in
the classification process. We then apply feature
selection techniques, such as information gain
and chi-squared test, to select the most
informative features for training the decision
tree. Next, we split the dataset into training,
validation, and testing sets using a stratified
sampling technique to ensure that the
distribution of classes is consistent across all
sets. We use the training set to train the decision
tree and tune the hyperparameters using cross-
validation and grid search to achieve the best
possible performance. To evaluate the
effectiveness of our proposed system, we use a
range of performance metrics, such as accuracy,
precision, recall, F1 score, and area under the
receiver operating characteristic curve
(AUCROC). We also conduct a comparative
analysis with other state-of-the-art machine-
learning algorithms to demonstrate the
superiority of our proposed system. Overall, our
proposed system provides a robust and effective
solution for detecting malicious bots on Twitter,
which can help prevent the spread of
misinformation, identify coordinated campaigns,
and protect the integrity of online conversations.

p-ISSN: 2395-0072
4. SYSTEM CLASSIFICATION
Fig -2: Architecture Diagram: Malicious bot diagram
To classify whether a bot on Twitter is
malicious or not, we can use various
supervised classification techniques like
Random Forest, Decision Tree, SVM, Naive-
Bayes, and K-Nearest Neighbors (KNN).
Here are the steps to implement the system
classification for malicious bot detection
using these techniques:
1. Data collection and preprocessing:
Collect a large dataset of tweets posted
by different Twitter accounts. Use the
Twitter API to extract tweets from
both bot and human accounts.
Preprocess the data by cleaning and
filtering the tweets, removing URLs,
mentions, stop words, and other noise
from the data.
2. Feature extraction: Extract relevant
features from the preprocessed tweet
data that can be used totrain the
classifiers. These features include
metadata such as the account age,
number of followers, and number of
tweets posted. They can also include
linguistic features such as the
sentiment, frequency of certain words,
and the presence of hashtags or
mentions.
3. Labeling: Label the dataset as either
malicious or non-malicious based on
whether the account has been identified
as a bot spreading spam or other
harmful content.
4. Splitting the dataset: Split the labeled
dataset into training and testing
datasets. Use the training dataset to
train the classifiers and use the testing
dataset to evaluate their performance.
5. Classification techniques: Use the
following classification techniques to
train and evaluate the classifiers:
 Random Forest: A random forest is
an ensemble of decision trees, where
each tree is trained on a subset of
the features and data. It can handle
large datasets and avoid overfitting.
 Decision Tree: A decision tree is a
simple and intuitive algorithm that
works by recursively splitting the
data into subsets based on the most
informative features.
 SVM: A Support Vector Machine is a
powerful algorithm for classification,
which works by finding a
hyperplane that separates the data
into different classes.
 Naive-Bayes: A Naive-Bayes
classifier works by assuming that
the presence of a particular feature
in a class is independent of the
presence of other features. It is a
simple yet effective algorithm for
text classification.
 KNN: A K-Nearest Neighbors
classifier works by finding the k-
nearest neighbors to the test
instance and classifying it based on
the majority class of its neighbors.
6. Model evaluation: Use metrics such as
accuracy, precision, recall, and F1-score
to evaluate the performance of each
classifier. Compare the results of each
technique and choose the one with the
best performance.

p-ISSN: 2395-0072
By following these steps, you can implement a
system classification for malicious bot detection
using ML-supervised classification techniques like
the random forest, decision tree, SVM, Naive
Bayes, and KNN for the Twitter dataset.
5. PERFORMANCE ASSESMENTS
5.1 Validations
 K-fold cross-validation: This technique
involves splitting the dataset into k-folds and
training the model on k-1 folds while testing
it on the remaining fold. This process is
repeated 'k' times. Cross validation can help
to reduce the risk of overfitting the model to
the training data.
 Stratified sampling: This is a technique where
the dataset is divided into homogeneous
subgroups (strata) based on the target
variable. This ensures that the training and
testing datasets have the same distribution of
target variables.
 Holdout validation: In this technique, the
dataset is divided into a training set and a
validation set.
5.2 Evaluations
To evaluate our system's effectiveness in detecting
malicious bots on Twitter using machine learning
classification, performance measurements are
necessary. Performance metrics provide information
on the performance of models, algorithms, or
processes being evaluated. We used various
indicators to evaluate the proposed model's
effectiveness, including a two-dimensional matrix
representing the actual and assigned class. This
matrix includes true positives, false positives, true
negatives, and false negatives to describe its
composition.
Table -2: Confusion Matrix Evaluation
Predicated
Bot
Predicted
Human
Actual Bot TP FN
Actual Human FP T N
The TP is a metric that indicates the successful
detection of inappropriate tweets from fake
Twitter accounts, while an FP refers to the
detection of inappropriate tweets that are not
associated with fake Twitter accounts, which
is known as a type 1 error. On the other hand,
an FN is when the detection system fails to
identify inappropriate tweets from fake
Twitter accounts, known as a type 2 error. TN
is when the detection system correctly
disregards inappropriate tweets that are not
associated with fake Twitter accounts. The
accuracy of the detection system can be
evaluated using a confusion matrix. The
accuracy (A) is a measure of how well the
detection operation aligns with the actual
value, which can be calculated using the error
rate. The error rate is the difference between
the detection rate and the current rate, divided
by the current rate, and expressed as a
percentage. The accuracy is calculated as the
sum of TP, TN, and NP, divided by the total
number of tweets.
Error rate=detection rate−current rate current
rate×100
Other performance measures include
precision/sensitivity, recall, sensitivity, and
specificity. Sensitivity is calculated as TP
divided by the sum of TP and FN, while
specificity is calculated as TN divided by the
sum of TN and FP. The detection rate is
calculated as TP divided by the sum of TP, TN,
FP, and FN. These metrics can be used to
evaluate the performance of the detection
system for identifying inappropriate tweets
from fake Twitter accounts.
6. RESULT AND ANALYSIS
The experimental analysis involved loading the raw
dataset into Python and using the NLTK module to
narrow down the focus to essential features, such as
"Lexical Diversity of Hate Comments / Biased
comments", "most Frequently Used Words in
Titles", "Punctuation", and "the Text Length". NaN
value columns were removed to clean up the
missing data. The Scikit-learn library was used to
construct the machine learning model, and
decisions on partitioning or splitting the data were
informed by the dataset's characteristics. Results
showing the accuracy, precision, recall, and/or F1
score of each algorithm on the training and test
datasets were used to quantify the performance of

p-ISSN: 2395-0072
each algorithm and show how well they were able
to distinguish between malicious and non-
malicious bots. Observations were made about
specific features or attributes of the Twitter
dataset that were most relevant for detecting
malicious bots, such as certain types of tweets,
hashtags, or user behaviors.
Fig -3: Accuracy Metrics of Classification
Algorithms
The decision tree algorithm has been chosen as
the "best" algorithm not only because it had the
highest accuracy, but also because it was
relatively easy to understand and explain
compared to the other algorithms. Decision trees
can be used to classify data that contains both
categorical and numerical features. They can also
handle missing values and noisy data, which can
be useful in real-world applications. Decision
trees can capture non-linear relationships
between variables, which means they can be used
to classify data that is not linearly separable. This
makes them a powerful tool for classification
problems in which there is no clear separation
between classes. Decision trees can be used as
part of an ensemble of algorithms, such as random
forests or gradient boosting, which can improve
their performance and accuracy. The user
interface (UI) of our classification system allows
users to easily upload, clean, and display Twitter
data, as well as select and evaluate various
machine learning algorithms for malicious bot
detection.
Fig -4: Home Screen
The UI consists of four main screens: the data
upload screen, the data cleaning screen, the
algorithm screen, and the results screen. The
data upload screen allows users to upload CSV
files containing Twitter data, such as user
account information, tweet content, and
engagement metrics.
Fig -4: Upload dataset
Once the data is uploaded, users can navigate
to the show dataset screen to view all the data
and then preprocess the data for analysis.
Fig -5: Show dataset screen

p-ISSN: 2395-0072
Once the data is uploaded, users can navigate
to the data cleaning screen to remove any
irrelevant or duplicate data and preprocess
the remaining data for analysis. On the
algorithm screen users can view from a range
of supervised classification algorithms,
including Naive Bayes, K-Nearest Neighbors
(KNN), Decision Tree, Random Forest, and
Support Vector Machines (SVM), as well as a
custom algorithm. Users can evaluate the
performance of the algorithm on the test
dataset on the results screen The results
screen shows various performance metrics,
such as accuracy, precision, recall, and F1
score, as well as confusion matrices and ROC
curves. In our experiments, we found that the
Decision Tree algorithm achieved the highest
accuracy.
7. FUTURE SCOPE
As bots become more advanced and
sophisticated, it is important to continuously
develop and improve detection methods. One
potential area for future research is the
integration of multiple detection techniques,
such as signature-based, behavior-based, and
anomaly detection methods. By combining
these techniques, the overall accuracy and
effectiveness of bot detection can be
improved. Another area for future research is
the exploration of new features and attributes
that can improve the accuracy of bot
detection. For example, the use of natural
language processing techniques can help
identify subtle linguistic differences between
bot and human-generated content.
Furthermore, the development of more
advanced machine learning algorithms, such
as deep learning, can improve the accuracy
and efficiency of bot detection. This can be
especially beneficial in processing large
amounts of social media data in real-time.
There are several other techniques that can
be used for malicious bot detection in
addition to supervised classification
techniques. Some of these techniques include:
1. Network-based detection: Network-
based detection techniques can be
used to identify bot activity based on
network traffic analysis. These
techniques can help detect bots that
may be using sophisticated evasion
techniques.
2. Hybrid approaches: Hybrid approaches
that combine multiple detection
techniques, such as supervised and
unsupervised learning, can improve
the accuracy and effectiveness of bot
detection.
3. Behavioral analysis: Behavioral
analysis techniques can be used to
identify bot activity based on patterns
of behavior that are indicative of bot
activity. Examples of behavioral
analysis techniques include analysis of
account creation patterns, post
frequency, and engagement rates.
Overall, the choice of technique depends on the
specific characteristics of the problem at hand
and the data available for analysis. A
combination of multiple techniques may be
necessary to effectively detect malicious bots on
social media platforms.
8. CONCLUSIONS
In this study, observed the detection and elimination
of bot activities are crucial for network security
management across various industries. Traditional
bot detection methods have limitations, but a novel
approach based on machine learning and supervised
learning algorithms has been developed to overcome
these challenges. This methodology offers versatility
in detecting any type of bot, making it effective in the
rapidly evolving landscape of bot technologies.
Overall, the use of machine learning and supervised
learning algorithms is a promising approach to bot
detection, offering an effective means of identifying
and mitigating potential bot activities across social
media platforms and beyond.
REFERENCES
[1] M.Sahlabadi,R.C.Muniyandi andZ.Shukur- Detecting
abnormal geste in social network Websites by using
a process mining- fashion-J.Comput.Sci. -vol. 10, no.
3, pp. 393- 402, 2014.

p-ISSN: 2395-0072
[2] M.Al- Qurishi, M.S.Hossain, M.Alrubaian,
S.M.M.Rahman andA.Alamri- using analysis of
stoner geste to identify vicious conditioning in
large-scale social networks-
IEEETrans.Ind.Informat., vol. 14, no. 2, pp. 799-
813, Feb. 2018
[3] Phillips. Efthimion1, Scott Payne1, Nick Proferes2,
‖ Supervised Machine Learning Bot Discovery
ways to Identify Social Twitter Bots ‖, Master of
Science in Data Science, Southern Methodist
University, 6425 Boaz Lane, Dallas, TX 75205,
2018.S.Barbon,Jr.G.F.C.Campos,G.M.Tavares,R.A.Iga
wa,M.L.Proen ¸ ca, Jr andR.C.Guido.
[4] Discovery of humans, licit bots, and vicious bots in
online social networks grounded on ripples, ACM
Trans. Multimedia Comput, Commun, Appl, vol. 14,
no. 1s, Feb. 2018, Art.no. 26
[5] F. Morstatter,L.Wu,T.H.Nazer,K.M.Carley andH. Liu,
A new approach to bot discovery Striking the
balance between perfection and recall, in Proc.
IEEE/ ACM Int. Conf.Adv. Social Network. Anal.
Mining, San Francisco, CA, USA,Aug. 2016, pp. 533-
540.
[6] Y. Zhou et al ProGuard Detecting vicious accounts
in a social network- grounded online elevations,
IEEE Access, vol. 5, pp. 1990- 1999, 2017.
[7] Xuan Dau Hoang and Quynh Chi Nguyen, Botnet
Detection Grounded On Machine Learning Ways
Using DNS Query Data.
[8] C.K.Chang, Situation analytics- A foundation for a
new software engineering paradigm, Computer,
vol. 49, no.1, pp. 24-33, Jan. 2016. 38 39
[9] Efthimion, Phillip George, Payne, Scott and
Proferes, Nicholas( 2018) ‖ Supervised Machine
Learning Bot Discovery Ways to Identify Social
Twitter Bots, ‖ SMU Data Science ReviewVol. 1No.
2, Composition 5.
[10] Ranjana Battur, Nagaratna Yaligar ‖ Twitter Bot
Discovery using Machine Learning Algorithms, ‖
International Journal of Science and Research(
IJSR) ISSN 2319- 7064 ResearchGate Impact
Factor( 2018)0.28 — SJIF( 2018)7.426
[11] C. Cai, L. Li, and D. Zengi, Behavior enhanced deep
bot discovery in social media, ‖ in Proc. IEEE Int.
Conf. Intell. Secur. Inform. ( ISI), Beijing, China, Jul.
2017, pp. 128- 130.

Detecting Malicious Bots in Social Media Accounts Using Machine Learning Technology.

More Related Content

Similar to Detecting Malicious Bots in Social Media Accounts Using Machine Learning Technology. (20)

More from IRJET Journal (20)

Recently uploaded (20)

Detecting Malicious Bots in Social Media Accounts Using Machine Learning Technology.