Fake Profile Detection on Social Networking Websites using Machine Learning | Python IEEE Project

Summary

The video presents a Python project focused on detecting fake profiles on social networking sites, specifically Instagram, using machine learning. It introduces the project's foundation, based on a 2023 conference paper, and describes enhancements, including the use of Random Forest and Decision Tree classifiers. The project achieves high accuracy, with Random Forest showing better results. The dataset consists of 576 records with 12 features. The video walks through the project execution, from dataset upload to model training, prediction, and performance analysis, ultimately demonstrating fake and real account detection.

Takeaways

• 🌐 The project focuses on detecting fake profiles on social networking websites, particularly Instagram, using machine learning techniques.
• 🔎 The system aims to identify fake accounts that may be used for fraudulent activities, cyberbullying, or other malicious purposes.
• 📈 The project uses two machine learning models: the Random Forest classifier and the Decision Tree classifier, with the former achieving higher accuracy.
• 📊 The dataset used for training and testing the models contains 576 records with 12 distinct features, such as profile picture, username, and number of followers.
• 🏁 The Random Forest classifier model achieved a training score of 100% and a test score of 93%, outperforming the Decision Tree classifier.
• 📝 The project is implemented in Python, using Flask for the web framework, and HTML, CSS, and JavaScript for the front end.
• 💻 The system architecture includes data preprocessing, feature selection, model application, and performance analysis.
• 📋 The project's user interface allows users to upload a dataset, preview it, and then train the models to predict whether an account is fake or real.
• 📈 The performance analysis section provides detailed metrics like recall, precision, F1 score, and confusion matrices for both classifiers.
• 📊 The project includes static charts for visualizing the accuracy comparison between the two models and the distribution of fake and real accounts in the dataset.

Q & A

What is the main focus of the project discussed in the video?

The project focuses on detecting fake profiles on social networking websites, specifically Instagram, using machine learning algorithms such as Random Forest and Decision Tree classifiers.

Which machine learning algorithms are used in the proposed project?

The proposed project uses two machine learning algorithms: Random Forest Classifier and Decision Tree Classifier.

How does the proposed project differ from the base paper?

While the base paper uses the SG Boost algorithm and does not focus on a specific platform, the proposed project enhances the system by targeting Instagram specifically and uses Random Forest and Decision Tree classifiers instead.

What are the accuracy scores achieved by the Random Forest and Decision Tree models in the project?

The Random Forest model achieved a training score of 100% and a test score of 93%, while the Decision Tree model achieved a training score of 92% and a test score of 92%.

What kind of dataset is used in the project?

The dataset used in the project contains 576 records with 12 distinct features, such as profile picture status, length of the username, number of posts, number of followers, and whether the account is labeled as fake or real.

What are some of the key features of the dataset used for training the models?

Key features of the dataset include profile picture status, ratio of numbers in the username, length of the full name, description length, external URL status, account privacy status, number of posts, number of followers, and number of followings.

What are the main advantages of the proposed system compared to the existing system?

The main advantages include focusing specifically on Instagram, using more effective machine learning algorithms (Random Forest and Decision Tree), and achieving higher accuracy in detecting fake accounts compared to the base system that uses SG Boost.

What software and tools are used to develop the project?

The project is developed using Python 3.10.9, the Flask web framework, and front-end technologies like HTML, CSS, and JavaScript.

What is the purpose of the performance analysis section in the project?

The performance analysis section compares the precision, recall, F1 score, and confusion matrix of both the Random Forest and Decision Tree models, highlighting their effectiveness in detecting fake accounts.

How does the project execute the detection process after setting up the environment?

After setting up the environment, the user runs the source code, uploads the dataset, trains the models, and then inputs specific account details to predict whether an account is fake or real using the trained machine learning models.

Outlines

00:00

🔍 Introduction to Fake Profile Detection Using Machine Learning

The video introduces a Python project focused on detecting fake profiles on social networking websites using machine learning. The project is based on a 2023 conference paper that initially proposed using the SGBoost algorithm for classification. However, the video’s project aims to enhance the paper’s system by focusing specifically on Instagram, using Random Forest and Decision Tree classifiers. It explains that fake profiles are often used for malicious activities, and outlines the accuracy of both models, with Random Forest outperforming Decision Tree.

05:01

⚙️ System Requirements and Project Setup

This section covers the technical setup of the project, including system and software requirements. The project uses Python 3.10.9, Flask for web development, and HTML, CSS, and JavaScript for the front end. Viewers are instructed to ensure the required Python libraries are installed before running the code. The process of executing the project is also detailed, from navigating to the source code directory to running the app and accessing the project through a browser.

10:02

📊 Data Set, Model Training, and Fake Account Detection

This part explains the dataset used, which contains 576 records with 12 features such as profile picture, username, description, and followers. The dataset is used to train the models to detect fake accounts. Two models are tested: Random Forest and Decision Tree classifiers. Viewers are shown how to upload the dataset, train the models, and make predictions on whether a given Instagram account is fake or real based on several features. Both models predict similar results, with Random Forest being slightly more accurate.

15:05

📈 Performance Analysis and Chart Comparisons

The performance of both models (Random Forest and Decision Tree) is analyzed, showing metrics such as recall, precision, and F1 scores. Confusion matrices are also presented to demonstrate the performance of each model in classifying accounts as real or fake. The section concludes with a comparison of the accuracy of both models, with Random Forest achieving 93% accuracy and Decision Tree 92%. Additionally, a chart compares the percentage of fake and real accounts in the dataset, which is composed of 60% fake and 40% real accounts.

Keywords

💡Fake Profile Detection

Fake profile detection refers to identifying fraudulent or false accounts on social networking websites. In the video, the project focuses on detecting fake Instagram accounts using machine learning models. This is crucial in combating cyber threats like scams, bullying, and identity fraud.

💡Machine Learning

Machine learning involves algorithms and statistical models that allow computers to make predictions or decisions based on data. In the video, machine learning models like Random Forest and Decision Tree classifiers are used to detect fake Instagram accounts by analyzing various features of user profiles.

💡Random Forest Classifier

Random Forest is a machine learning algorithm that builds multiple decision trees and merges their results to improve accuracy. In the video, it is one of the models used for fake profile detection and has achieved a training score of 100% and a test score of 93%, making it the more effective of the two models.

💡Decision Tree Classifier

A Decision Tree Classifier is a model that splits data into branches to predict outcomes. The video describes using this model for fake account detection with a training and test score of 92%. While effective, it is slightly less accurate than the Random Forest model in this project.

💡Instagram Fake Account Detection

Instagram Fake Account Detection is the focus of the project in the video. It refers to identifying fraudulent accounts specifically on the Instagram platform, using machine learning techniques to analyze user behavior and profile details like username, posts, and followers.

💡SGBoost Algorithm

SGBoost (Stochastic Gradient Boosting) is an algorithm that enhances the performance of machine learning models by focusing on difficult-to-predict cases. The video mentions that the base paper used SGBoost, but the project in the video opts for Random Forest and Decision Tree classifiers instead for improved performance.

💡Dataset

A dataset is a collection of data used for training machine learning models. The video mentions a dataset with 576 records and 12 features, including profile pictures, username length, and the number of followers. This data is used to train the model to predict whether an account is fake or real.

💡Profile Features

Profile features refer to characteristics of a social media profile used for machine learning analysis. In the video, these include attributes like the number of followers, the length of usernames, the presence of an external URL, and whether the account is private. These features help determine the likelihood that an account is fake.

💡Accuracy

Accuracy in machine learning refers to the percentage of correct predictions made by the model. In the video, the accuracy scores of the Random Forest (93%) and Decision Tree (92%) classifiers are highlighted, indicating how well each model performs in detecting fake Instagram accounts.

💡Performance Analysis

Performance analysis evaluates how well a machine learning model performs by analyzing metrics like precision, recall, and F1 score. The video shows a comparison of performance for both the Random Forest and Decision Tree classifiers, demonstrating that Random Forest slightly outperforms the other model.

Highlights

Introduction to the Python project on fake profile detection using machine learning.

Discussion on the rising usage of social networking sites and issues like fake accounts.

Base paper titled 'Fake Profile Detection on Social Networking Websites using Machine Learning' utilizing SGBoost algorithm.

The enhanced project focuses on Instagram fake account detection using machine learning.

Two models implemented: Random Forest classifier and Decision Tree classifier.

Random Forest classifier achieved a training score of 100% and a test score of 93%.

Decision Tree classifier achieved both training and test scores of 92%.

Dataset contains 576 records with 12 distinct features like profile pic, username length, and followers.

Comparison of the enhanced system with the base paper, highlighting improvements.

Step-by-step instructions for executing the Python project in a command prompt.

A static login page is used, without any database integration.

Prediction model tests for Instagram accounts with both Random Forest and Decision Tree classifiers.

Performance analysis showing precision, recall, and F1 scores for both models.

A static chart comparing accuracy scores of Random Forest (93%) and Decision Tree (92%).

Final visualization of the dataset showing a 60% fake and 40% real account ratio.

Transcripts

[Music]

hi in this video we going to see about a

python project which is entitled as fake

profile detection on social networking

websites using machine learning which is

an i 2023 conference paper before seeing

the execution of the project let me

leave about this project so as we know

that the internet is growing everywhere

all over the world the same way the

usage of social networking is also

happens all over the world and all over

the people the people come to social

networking sites to spend their time for

various reasons but some of the fake

users will be creating some frake

accounts to sell

some products or cheat some people for

making money or they may threaten or

make cyber bullying so there are various

reasons for creating this kinds of fake

profiles so here in the base paper the

authors have proposed a system for

detecting the fake profile detection on

social networking website using machine

learning and they have used SG

boost algorithm for making the

classification and here in the base

paper we not going to implement the same

as mentioned in the base paper so we are

going to enhance some features other

than the base paper so here the

drawbacks that is mentioned in the base

paper is like they have prescribed about

the social networking website but they

are not prescribed about any particular

website so they are generally mention

about the social networking websites

only so now we are going to enhance the

system so now let us see about other

enhancement so here you can see the I

base paper title is fake profile

eduction on social networking website

using machine learning or our proposed

project title is Instagram fake account

detection using machine learning so we

are going to concern about the Instagram

social networking website only and here

you can see the it base paper abstract

let us see about the proposed abstract

so here in the proposed system so we'll

be implementing two different

models so the we have used two different

models for our proposed systems the

first model using random Forest

classifier and the second model using

the decision tree classifier so the

first model that is using random Forest

classifier we have achieved the train

score of 100% and test score of

93% and the second model of dentry

classifier we have achieved train score

of 92% and death score of 92% so from

the two models we'll be proving that the

random Forest classifier is performing

well on it so now coming to the coming

back to the abstract part so here you

can see the

abstract that we have mentioned about

the Instagram fake account detection is

in machine learning using python we have

done that and we have used the two

distinct algorithms like random forest

classifier and dentry so here you can

see the accuracy of the models and

coming to the data set part so we have

used the data set which contains 576

data set records and there will be 12

distinct features on it so there are

various features on it so I'll show you

the data set part now so this is the

data set that we have used so as

mentioned you can see scroll down and

you can see which contains around 576

data center records and here you can see

the two 12 uh distinct features like

profile

pick numbers by length of the user full

name words numbers by length of full

name name is equal to username

description length external URL private

number of post number of followers

number of followers and fake that is

labeled as Z or one so these all this is

the data set that you're going to train

up the system coming back to the

abstract document so here as mentioned

so we'll be using Python and we'll be uh

finding out the account is fake or not

this is the about the existing system so

we are considering the base paper as

existing so as mentioned the base paper

they are used SG boost algorithm so we

have described about the existing system

part here and coming to the next next

part that is the disadvantages of

existing system so we have listed the

disadvantages of the existing system

that is using the SG boost and coming to

the propose system so we have mentioned

about the propose system what we have

done in

it and next part is the advantages of

the propos system so these are the

advantages that we have listed of about

our propose system the system

architecture you can find that input

data set and we are going to make the

free processing and feature selection

and we have app apply the random forest

classifier and add classifier the

predicted result is it is a fake or real

account and we'll be showing the

performance analysis and the graph part

of it in the system requirements you can

see the hardware and software

requirements as mentioned we have

developed using python the version that

we have used is Python

3.10.9 and web framework is flask and

the front end part we have done using

HTML Cs and JavaScript and this is the

reference of the project is the base

paper of the

project

before execution make sure that you have

fulfilled the requirement that is

mentioned the reement file with the

exact verion of the Python and the

library is installed in your system now

let us see the execution of the project

so just go to the source code location

copy the source code location now go to

the command

[Music]

prompt now go to the drive location

where you have pasted the code in my

case I have pasted my code in F drive

I'll go to the F drive now now type CD

and give space and paste the location

that we are copied and click enter so

now we are into the source code location

now type Python app.py and click enter

and kindly wait for a few

seconds so now you can see the URL just

copy this URL go to any of your browser

I'm going to Google Chrome now paste the

URL that we are copied and click enter

so now you can see the home screen or

welcome screen of the project with the

project title Instagram fake account

deduction using machine learning

so first click this login

menu so once if you click this login

menu it will be navigated to the login

page kindly note that this is a static

login page because we have not used any

database in the project so just enter

the default username and password as

admin and admin and then click the login

button now you can see the login success

message and click okay so now it will be

navigated to the upload part where you

need to upload the data set just select

this choose file now go to the source

code location

and select the upload. CSV file and then

click this upload

button so once I've uploaded the page

will be navigated to the preview page so

where you can preview the data set that

we have uploaded so as

mentioned shown earlier you can see the

features like ID profile pick number

length user full name words so all the

12 features you can see here and if you

scroll down till the end you can find

the complete data set has been loaded

into the pre view part with the all the

575 data set records now you can just

click this click to train not test

button and kindly wait for few

seconds so now you can see the training

finish message and click okay so now it

will be navigated to the important part

that is the prediction part so where you

need to enter the details and check

whether the prediction result with the

account is a fake or not so here you can

see the two models that is model with

random forest classifier and d three

classifier so you can check with the any

any one model by selecting the model and

I'll show you with the few cases now so

now let me enter the profile pick as

yes ratio of numbers by length username

is uh

0.55 full name words as one ratio of

numbers full length full name is

0.44 and username is equal to name is

equal to username no description length

is

zero external URL is

no account private is no total number of

post is 33 total number of followers is

166 and total follows is 59

56 and here you can select the model as

mentioned I I'll show you with the

random forest classifier and click this

predict

button so now you can see the prediction

result is the account is the Instagram

account is it is a fake account so what

are the details that is mentioned that

is a fake account the model that is we

have used is random Forest classifier so

I'll show you with the as we are not

using any database the values that we

have entered has been reset now so I

I'll just enter the same values again

and I'll show you with the other model

also

[Music]

[Music]

quickly so now let me select the model

as decision tree classifier and click

the predict button so now you can see

the addition Tre model also is

classified this account and predicted

the result is all is a fake account so

in this way you can just check with both

the models uh

whether random for us at D Tre maximum

both will be showing the same result as

we got only the few difference

percentage in the accuracy so let me

show you with the other case now with

the profile picture

yes now ratio of numbers

zero full name words

two ratio of numbers name Z

zero user name is equal to usern name no

description length is

63

external URL is

[Music]

no account private is no total number of

posts

is

378 total number of followers is

34

670 total follows is

1,878 so now let us check with the model

random for us and click the predict

button and now you can see the model

predicted the Instagram account is real

so I shown you with a a case of real now

for examp example I'll take an account

of ours so let me type JP infotech

Instagram let me take this

example so

here if you wanted to check again just

click this prediction menu again so now

it is reseted now we are in the

prediction menu so now the profile

picture is available so I'll click

profile picture yes so ratio of numbers

I have not used any numbers in the

username so it is zero f full name words

so full name words we have three words

as full name words so here also we don't

have a number so let me give zero

username name is equal to username so I

have the username and the name is same

so I'll give it as yes description

length so description length will be

around

150 characters maybe I'll just enter

approximate of 150 characters and uh

external URL yes I have given external

URL JP info.org so I'll just enter that

yes and account is private no this is

not a private account this is a public

account so I'll click yes total number

of post I have posted is one1 post so

I'll just enter1 and total followers is

275 followers so I just enter 275

followers and following zero I just

enter zero so now let us check for this

case and click the predict

button and now you can see the model

random Forest classified this Instagram

account is a real one so in this

scenario you can check with uh some

other cases also so now let me come to

the prediction menu again so now let me

enter the case with profile picture

yes ratio number

0.07 full name words as one there are

zero other one is no description length

are

zero external URL no private as yes

total number of poost are zero total

number of followers is 47 and total

follows is 98 and let me click the

predict so now you can see this account

is predicted as a fake account so in

this way you can check with the other

cases so this is not limited only to the

things that I have mentioned there are

as I mentioned you there are around uh

500 data set records you can check with

each and every one with both the models

and you can just find the results of it

so now to make the video shorter I move

to the next part that is the performance

analysis part so just click this

performance

analysis so now you can see the

performance analysis of both the models

That Is Random forest classifier and for

the dentry classifier so in the

performance analysis of random Forest

classifier you can see the recall

Precision in F1 score so the recall

value Precision value and F1 score for

the both the cases 0o and one has been

shown and here you can see the confusion

Matrix of the random Forest classifier

which contain two one print label of the

both the cases 0 and one and coming to

the next model that is dision Tre

classify performance analysis you can

see the recall PR and F1 score of this

di Tre classifier for both cases 0 and

one and you can see the confusion Matrix

for the dentry classifier which contains

the two and PR label for the both

cases and finally is the chart part so

just click this chart chart menu to be n

the chart but kindly note that this

chart is also a static chart because we

are not used any database in the project

so first chart shows the comparison of

the accuracy score so here you can see

we have used two different model like

random foration Tre classifier So Random

for us has achieved the accuracy of 93%

and tree of about 92% so that has been

compared here and we have proved that

our random fors performs better well

than other model and coming to the next

chart which contains

the uh the fake and real percentage that

is the data set the which we have

trained up with contains the fake

account of 60% and real account of 40%

data set record so that is being

depicted here manually so that is what I

said this is a static chart so the chart

part con sometimes the comparison of

accuracy score and the data set

comparison and now let me

loog and this is all about the project

Instagram fake account detection using

machine learning using Python and thank

you for watching