PROJECT BASE LEARNING

Faculty of Engineering and Technology

ITM Vocational University

1
 FAKE NEWS DETECTION

Submitted
By

Shingote Amruta 160110424044

Anjali Parmar 160110423006

A Project Report Submitted to

ITM Vocational University in Partial Fulfillment of the Requirements for
The Degree of [Link] in
Computer Science Engineering Department

May 2026

CSE Department
ITM Vocational University,
At & PO – Raval – Waghodia,
Ajwa Nimeta Road,
Dist. Vadodara – 391760
Telephone No. 2668-292840

2
 DECLARATION

Project Title: STOKE MARKET ANALYSIS

We hereby declare that the project report entitled “Fake News Detection”,
submitted in partial fulfillment of the requirements of the Project Based
Learning (PBL) module at ITM Vocational University, Vadodara, is the result of
our own work and effort.

The project has been completed under the guidance of our faculty and has not
been submitted elsewhere for the award of any other degree, diploma, or
certificate. All the information and data provided in this report are true to the
best of our knowledge and belief.

We understand that any form of plagiarism or violation of academic integrity

will lead to disciplinary action as per the university policies.

Submitted by:
Name Enrollment No.
Shingote Amruta 160110424044
Anjali Parmar 160110423006

Date: 29/5/2026

Signatures: _______________________

3
 CERTIFICATE

It is certified that the work contained in this project report entitled “Stock Market
Analysis” submitted by Shingote Amruta & Anjali Parmar students of ITM
Vocational University, for the award of the [Link] (Computer Science
Engineering) degree, is absolutely based on their own work carried out under my
supervision and that this work has not been submitted elsewhere for any degree.

Head of Department

Dr. Rajdipsinh Vaghela

Associate Professor
Computer Science Engineering,
ITM Vocational University

Date:
Place:

4
 COMPLIANCE CERTIFICATE
This is to certify that research work embodied in this project report entitled “Fake
News Detection” was carried out by:
 Shingote Amruta – 160110424044
 Anjali Parmar – 160110423006
at ITM Vocational University for partial fulfillment of the [Link] degree to be
awarded by ITM Vocational University. They have complied with the comments
of PROJECT – I to our satisfaction.

Date:
Place:

5
 PROJECT APPROVAL CERTIFICATE

This is to certify that project work embodied in this project report entitled “Stock
Market Analysis”, carried out by Shingote Amruta & Anjali Parmar at the
Faculty of Vocational Professional & Interdisciplinary Studies, is approved
for the degree of Bachelor of Computer Application in the CSE Department
by ITM Vocational University, Vadodara.

Date:
Place:

Examiners’ Sign and Name:

1) _________________________________________

2)_________________________________________

6
ABSTRACT
The rapid proliferation of fake news on social media
and digital platforms poses a significant threat to public
opinion and democratic processes. This project presents
an automated system for detecting fake news using
Natural Language Processing (NLP) and Machine
Learning. The proposed system analyzes textual data
from news articles, processes it using NLP techniques,
and classifies it as either "real" or "fake" with high
accuracy. The project includes a user-friendly web
interface that allows users to input news text or a URL
and receive an instant prediction, complete with a
confidence score and an explanation of the result. The
developed system successfully demonstrates that a
Passive Aggressive Classifier, combined with TF-IDF
vectorization, can serve as an efficient and scalable
solution for combating misinformation.

7
TABLE OF CONTENTS
1. Introduction
2. Literature Survey
3. System Requirements & Analysis
4. System Design
5. Implementation
6. Testing & Results
7. Conclusion and Future Scope
8. References

8
CHAPTER 1: INTRODUCTION

1.1 Overview of Fake News

In the digital age, information travels faster than ever
before. With the rise of social media platforms, blogs,
and online news portals, the dissemination of
information has become instantaneous. However, this
accessibility has also given rise to a significant
challenge: the proliferation of "fake news."
Fake news refers to deliberately false or misleading
information presented as legitimate news. It is often
created to influence political views, generate advertising
revenue through clickbait, or simply cause public
confusion. Unlike traditional misinformation, fake news
is typically fabricated with the intent to deceive.
The impact of fake news is profound. It can manipulate
public opinion during elections, incite social violence,
and even endanger public health by spreading medical
falsehoods. Distinguishing between credible journalism
and fabricated content has become increasingly difficult
for the average user, making the detection of fake news
a critical issue in today's society.

1.2 Problem Statement

The primary problem addressed in this project is the
inability of current systems to efficiently and accurately
identify fake news at scale. Manual fact-checking by
human journalists is the gold standard for verifying
information; however, it is a slow, labor-intensive
process that cannot keep up with the millions of articles
published daily.
Furthermore, many existing automated solutions are
9
either "black boxes" that provide no explanation for
their decisions, or they are too complex for the general
public to use. There is a pressing need for an
automated, intelligent system that can:
1. Analyze news articles in real-time.
2. Classify them as "Real" or "Fake" with high
accuracy.
3. Provide a user-friendly interface accessible to non-
technical users.

1.3 Objectives
The main objective of this project is to develop an
automated web-based application for the detection of
fake news using Natural Language Processing (NLP)
and Machine Learning techniques. The specific
objectives are:
1. Data Analysis: To collect and preprocess a large
dataset of news articles to understand the linguistic
patterns that distinguish fake news from real news.
2. Model Development: To implement and train various
machine learning algorithms (such as Logistic
Regression, Naive Bayes, and Passive Aggressive
Classifier) to identify the most accurate model for text
classification.
3. Feature Engineering: To convert raw text data into
numerical features using techniques like TF-IDF
(Term Frequency-Inverse Document Frequency) to
make it understandable for machine learning models.
4. Application Development: To build a user-friendly
web interface using Streamlit that allows users to
input news text or URLs and receive instant
predictions.
5. Explainability: To integrate features that highlight
suspicious words or phrases, providing transparency
to the user regarding why a specific article was
10
 flagged as fake.
1.4 Scope of the Project
The scope of this project is currently limited to text-
based news articles. The system analyzes the linguistic
patterns, metadata, and content of the text to make
predictions.
Limitations:
 Language: The model is currently trained on English

language datasets and may not perform accurately on

news in other languages.
 Satire: The system may struggle to distinguish

between malicious fake news and satirical content that

is intentionally humorous but factually incorrect.
 Media: This project focuses solely on text analysis; it

does not currently analyze images or videos for

manipulation (deepfakes).
Future Scope: Future enhancements could include
extending the model to support multiple languages,
integrating image analysis to detect manipulated media,
and developing a browser extension for real-time fact-
checking while browsing social media feeds.

11
CHAPTER 2: LITERATURE SURVEY

2.1 Introduction
The detection of fake news has garnered significant
attention from the research community in recent years.
Various approaches have been proposed, ranging from
simple linguistic pattern matching to complex deep
learning architectures. This chapter reviews existing
literature and related work to understand the evolution
of fake news detection techniques. By analyzing
previous studies, we can identify the strengths and
limitations of current methods and establish a
foundation for the proposed system.

2.2 Review of Existing Approaches

Researchers have approached the problem of fake news
detection primarily through two lenses: content-based
analysis (analyzing the text) and context-based analysis
(analyzing the user or propagation network). This
project focuses on content-based analysis. Below is a
review of key methodologies found in literature:

2.2.1 Detection Using N-grams and Naive Bayes In early

studies, researchers focused on linguistic features. One
prominent approach utilized the Naive Bayes classifier
with N-gram features (sequences of N words).
 Methodology: The study used a bag-of-words model

where the frequency of specific word pairs (bigrams)

or triplets (trigrams) was analyzed. Naive Bayes was
chosen for its speed and probabilistic nature.
 Findings: This method proved effective for spotting

obvious clickbait headlines that contained

sensationalist vocabulary. However, it struggled with
12
 complex sentence structures and sarcasm, often
failing to identify sophisticated fake news articles that
mimicked formal journalistic styles.

2.2.2 Hybrid Ensemble Learning Techniques To

improve accuracy, some researchers proposed hybrid
systems combining multiple machine learning
algorithms, such as Support Vector Machines (SVM)
and Random Forest.
 Methodology: In these studies, text was converted into

numerical vectors using TF-IDF. These vectors were

then fed into an ensemble of classifiers where a voting
mechanism decided the final label.
 Findings: The hybrid models achieved higher

accuracy (around 85-88%) than individual classifiers.

They were robust against noise in the data. However,
the training time increased significantly, making real-
time prediction challenging on low-end hardware.

2.2.3 Deep Learning with LSTM Networks With the rise

of deep learning, Recurrent Neural Networks (RNNs),
specifically Long Short-Term Memory (LSTM)
networks, became a popular choice.
 Methodology: LSTM networks are capable of

remembering long-term dependencies in text.

Researchers fed pre-processed word embeddings (like
GloVe or Word2Vec) into LSTM layers to capture the
sequential context of the news article.
 Findings: LSTMs outperformed traditional machine

learning models, achieving accuracy rates exceeding

90%. They successfully understood the context of a
sentence rather than just counting keywords.
However, these models require large amounts of
training data and substantial computational power
(GPUs) to converge.
13
 2.2.4 Transformer Models (BERT) More recently, state-
of-the-art results have been achieved using
Transformer-based models like BERT (Bidirectional
Encoder Representations from Transformers).
 Methodology: BERT uses a multi-head attention

mechanism to understand the relationship between all

words in a sentence simultaneously, regardless of
their distance from one another.
 Findings: BERT-based models currently hold the

highest benchmarks for fake news detection.

However, they are computationally expensive and
have high latency, making them difficult to deploy in
lightweight web applications without significant cloud
infrastructure.

2.3 Comparative Analysis

Training
Approach Accuracy Speed Pros
Simple, easy to
Moderate implement, low
Naive Bayes (70-80%) Very Fast resource usage.
Robust, handles
Ensemble Good (85- high-dimensional
(SVM/RF) 88%) Moderate data well.
Captures context
LSTM (Deep High (90- and sequential
Learning) 92%) Slow dependencies.
State-of-the-art
BERT Very High context
(Transformer) (94%+) Very Slow understanding.

14
2.4 Research Gap
While recent advancements in Deep Learning and
Transformers (like BERT) have achieved remarkable
accuracy, they come at the cost of high computational
complexity. Existing literature shows a gap in
developing lightweight models that are:
1. Accurate enough for general use.
2. Fast enough for real-time web applications without
requiring expensive GPU servers.
3. Explainable, providing users with reasons for the
prediction rather than a black-box output.
This project aims to bridge this gap by utilizing a
Passive Aggressive Classifier, which is known for its
efficiency with large datasets, combined with TF-IDF
vectorization. This approach balances accuracy with
speed, making it suitable for a scalable web application.

15
CHAPTER 3: SYSTEM REQUIREMENTS &
ANALYSIS

3.1 Introduction
Before designing and developing the software, it is
crucial to identify the resources required for the system
to function correctly. This chapter outlines the
hardware and software specifications necessary for the
development and execution of the Fake News Detection
System. Additionally, it details the functional and non-
functional requirements that define the system's
behavior and performance.

3.2 Hardware Requirements

The hardware requirements for this project are modest,
making it accessible on standard consumer-grade
laptops and desktops. However, since the project
involves training machine learning models on datasets,
a configuration with sufficient processing power and
memory is recommended.
 Processor: Intel Core i3 / i5 / i7 or AMD Ryzen 3 / 5 /

7 equivalent.
 RAM (Memory): Minimum 4GB RAM; 8GB

recommended.
 Storage: Minimum 10 GB of free hard disk space.

 Network: A stable internet connection (for setup).

3.3 Software Requirements

 Operating System: Windows 10/11, Ubuntu (Linux),

or macOS.
 Programming Language: Python 3.8 or higher.

 IDE: Visual Studio Code or Jupyter Notebook.

 Libraries: Pandas, NumPy, Scikit-learn, NLTK,

16
 Streamlit, Pickle.
3.4 Functional Requirements
 FR-01 (User Input): The system shall allow the user to

input news content via a text area.

 FR-02 (Preprocessing): The system must

automatically clean the input text before prediction.

 FR-03 (Prediction): The system shall classify the input

news as "Real" or "Fake."

 FR-04 (Confidence Score): The system shall display

the probability percentage.

3.5 Non-Functional Requirements

 NFR-01 (Performance): Prediction response time

should be less than 2 seconds.

 NFR-02 (Accuracy): The model should achieve a

minimum accuracy of 85%.

 NFR-03 (Usability): The user interface must be

intuitive.

3.6 Feasibility Study

 Technical Feasibility: All required technologies are

mature and well-documented.

 Economic Feasibility: All tools are open-source and

free.
 Operational Feasibility: The target audience can use

the system with minimal training.

17
CHAPTER 4: SYSTEM DESIGN

4.1 System Architecture

The system follows a modular architecture divided into
distinct layers:
1. Presentation Layer: Streamlit Web Interface.
2. Application Layer: Python Backend Logic.
3. Processing Layer: Text Preprocessing & TF-IDF
Vectorization.
4. Model Layer: Trained Passive Aggressive Classifier.
5. Data Layer: SQLite Database.

4.2 Data Flow Diagrams (DFD)

 Level 0 DFD: User inputs text -> System processes it -

> System returns prediction.

 Level 1 DFD: Input -> Preprocess -> Vectorize ->

Predict -> Output.

4.3 UML Diagrams

 Use Case Diagram: Actors: User, Admin. Use Cases:

Login, Predict, Retrain Model.

 Class Diagram: Classes: User, NewsProcessor,

Predictor, DatabaseManager.

4.4 Database Design

Table: prediction_history
Column Name Data Type Description
id INTEGER Primary Key
date_time DATETIME Timestamp of prediction
input_text TEXT News content entered by us

18
 Column Name Data Type Description
prediction VARCHAR Stores result as "Real" or

confidence FLOAT Probability/confidence scor

19
 CHAPTER 5: IMPLEMENTATION

5.1 Introduction
This chapter details the practical implementation of the
proposed system.

5.2 Module Implementation

5.2.1 Data Loading Module The dataset ([Link]) was

loaded using Pandas. Rows with missing values were
handled.
[INSERT CODE SNIPPET: Importing Pandas and
loading CSV]

5.2.2 Data Preprocessing Module Text was cleaned by

lowercasing, removing punctuation, removing
stopwords, and lemmatizing using NLTK.
[INSERT CODE SNIPPET: Text cleaning function]

5.2.3 Feature Extraction Module TF-IDF Vectorizer

was used to convert text into numerical vectors.
[INSERT CODE SNIPPET: TF-IDF Vectorization
code]

5.2.4 Model Training Module A Passive Aggressive

Classifier was trained on the vectorized data.
[INSERT CODE SNIPPET: Model training code]

5.2.5 Web Application (Streamlit) The frontend accepts

user input, loads the trained model using pickle, and
displays the result.
[INSERT CODE SNIPPET: Streamlit app code]

20
5.3 Implementation Challenges
Challenges included handling dimensionality
mismatches during vectorization and optimizing the
lemmatization process for speed.

21
CHAPTER 6: TESTING & RESULTS

6.1 Test Cases

Expected
Test Case ID Input Data Actual Output
Output

Fake News
TC_01 Fake Fake
Text

Real News
TC_02 Real Real
Text

Empty
TC_03 Error Error
Input

6.2 Model Performance Metrics

 Accuracy: [Insert %]

 Precision: [Insert %]

 Recall: [Insert %]

 F1-Score: [Insert Score]

[INSERT IMAGE: Confusion Matrix]

6.3 System Screenshots

[INSERT SCREENSHOT: Home Page] [INSERT
SCREENSHOT: Fake News Result] [INSERT
SCREENSHOT: Real News Result]

22
CHAPTER 7: CONCLUSION AND FUTURE SCOPE

7.1 Conclusion
The project successfully demonstrated an automated
system for fake news detection. The combination of
NLP techniques and the Passive Aggressive Classifier
proved effective, achieving high accuracy and fast
response times. The web interface makes the technology
accessible to a general audience.

7.2 Future Scope

1. Multilingual Support: Extending the model to other
languages.
2. Multimodal Analysis: Adding image/video deepfake
detection.
3. Browser Extension: Creating a Chrome extension for
real-time browsing.

23
8. REFERENCES

1. Kaggle Inc., "Fake News Dataset," [Link].

2. Scikit-learn: Machine Learning in Python, Pedregosa
et al., JMLR 12, 2011.
3. Streamlit Documentation, [Link]

24