A MINI PROJECT REPORT ON

EMAIL SPAM DETECTION

SUBMITTED IN THE PARTIAL FULFILLMENT OF THE REQUIREMENTS OF
UNIVERSITY OF MUMBAI FOR THE AWARD OF THE DEGREE

OF

BACHELOR OF ENGINEERING
IN
COMPUTER ENGINEERING
SUBMITTED BY

1. Meet Parmar 11
2. Vandana Tripathi 04
3. Aditya Thakur 01

UNDER THE GUIDANCE OF

Ms. Sanjana Satpute

DEPARTMENT OF ARTIFICIAL INTELLIGENCE AND

MACHINE LEARNING

VIDYA VIKAS TRUST’S

UNIVERSAL COLLEGE OF ENGINEERING
KAMAN ROAD, VASAI 401208
UNIVERSITY OF MUMBAI
2023 -2024

1
 CERTIFICATE

This is to certify that the mini project report entitles

“ EMAIL SPAM DETECTION”

Submitted by
1. Meet Parmar - 11
2. Vandana Tripathi – 04
3. Aditya Thakur - 01

is a bonafide work has been carried out by them under the supervision of Ms. Sanjana Satpute
and it is approved for the partial fulfillment of the requirement of Savitribai Phule Pune University,
for the award of the degree of Bachelor of Engineering ( Computer Engineering).

Ms. Sanjana Satpute Dr. Jitendra Saturwar

Guide Head Of
Department of COMPS Department of COMPS

(Dr. J. B. Patil)
PRINCIPAL,
VIDYA VIKAS TRUST’S
UNIVERSAL COLLEGE OF ENGINEERING
KAMAN ROAD, VASAI 401208

2
 ACKNOWLEDGEMENT

I hereby take this opportunity to record my sincere thanks and heartily gratitude to Ms. Sanjana
Satpute for his useful guidance and making available to me his intimate knowledge and experience
in making “EMAIL SPAM DETECTION” as a preparation of report in respect there of. I am
also thankful to my respective HOD Dr. Jitendra Saturwar of my Computer Engineering
department. I express my special thanks and heartily gratitude to my respective staff members for
inspiring me throughout the completion of this system. The acknowledge will be incomplete if I
don’t record sense of gratitude to my principal. I also express my sincere thanks to all those i.e.
the management, lab assistants, my friends and family who have provided me valuable guidance
towards the completion of this presentation as a part of the syllabus of the course.
I express my sincere gratitude towards co-operative department who have provided me
with valuable assistance and requirements for the presentation.

3
 CONTENTS

Sr. No Title of Chapter Page no.

Abstract 5
1. Introduction 6
2. Literature Review 7
3. Requirements 8
4. Methodology 10
5. Implementation 12
6. Results 14
7. Conclusion 16
8. References 16
9. Appendices 17

4
 ABSTRACT

In the modern digital age, email has become one of the most prevalent means of communication,
used across personal, professional, and academic spheres. However, with the increased use of
email has come an equally significant rise in spam emails, which can often be malicious and
deceptive in nature. Spam emails can contain phishing attempts, malware, advertisements, or
unsolicited content, posing threats to both individuals and organizations. The automatic detection
and filtering of these spam emails have become a crucial task in ensuring the security and
integrity of online communication.

Traditional spam detection techniques typically relied on keyword-based filtering and heuristic
rules, which are prone to false positives and can struggle to adapt to evolving spam tactics. To
address this challenge, machine learning and natural language processing (NLP) techniques have
emerged as powerful tools in building more intelligent and adaptive spam filters. These
techniques not only enable systems to learn from data but also improve over time as new types of
spam emerge.

This project leverages NLP techniques and machine learning algorithms to detect spam emails
automatically. By using Python and Jupyter Notebook, we developed a spam detection system
that processes raw email data, extracts key features from the text, and uses machine learning
models to classify emails as either spam or ham .The project makes use of various machine
learning algorithms, including the Naive Bayes classifier, known for its efficiency in text-based
classification, and the Support Vector Classifier (SVC), recognized for its high performance in
binary classification tasks. Additionally, the project implements text preprocessing steps such as
tokenization, stopword removal, and text vectorization to transform the email data into a
structured format that the machine learning models can utilize effectively.

Through this project, we aim to showcase the effectiveness of combining NLP and machine
learning techniques for spam detection. The models were trained on a widely used public dataset
containing thousands of labeled emails, with each email categorized as spam or ham. Our results
demonstrate that both Naive Bayes and SVC models can achieve high accuracy in classifying
spam emails, although each has its own strengths and trade-offs.

Overall, this project highlights the significance of machine learning in addressing real-world
challenges like email spam detection. It also emphasizes the importance of feature extraction,
data preprocessing, and model selection in achieving accurate classification results. This spam
detection system could potentially be deployed in real-world email platforms to enhance their
ability to filter out malicious or irrelevant content, thus safeguarding users from potential threats
and improving email communication security.

5
 CHAPTER 1

INTRODUCTION

1.1 Brief Overview

Email communication has become an integral part of personal and professional lives, but it is often
inundated with spam, which includes unsolicited advertisements, phishing attempts, and other
malicious content. With the rising volume of email traffic, distinguishing between legitimate and
spam emails has become a crucial challenge. This project addresses this issue by implementing a
machine learning-based spam detection model. Using a dataset of labeled emails, the project
applies NLP techniques and classification algorithms to build a robust detection system.

1.2 Importance and Relevance

Spam detection is of significant importance for individuals and organizations, as spam emails can
lead to loss of time, productivity, and even data breaches through phishing attacks. Automated
spam detection systems help to filter out such unwanted messages, allowing users to focus on
legitimate content. The relevance of this project lies in its ability to enhance cybersecurity,
safeguard sensitive information, and improve overall email communication efficiency.1.3
Objectives of the Study

The main objectives of this project are:

 To develop a spam detection system that can classify emails as spam or

legitimate based on their content.
 To leverage NLP for text preprocessing to improve the accuracy of machine
learning models.
 To implement and compare the performance of two widely used classifiers,
Naive Bayes and SVC, for spam detection.
 To provide visual insights into the distribution of spam and legitimate emails
using data visualization techniques.

6
2. Literature Review

2.1 Summary of Existing Work

Over the past two decades, machine learning has emerged as one of the most effective tools
for spam detection. By automatically learning from large datasets, machine learning
models can generalize patterns found in spam emails and apply them to unseen emails.
Several studies have demonstrated the efficacy of various machine learning algorithms,
including Naive Bayes, Support Vector Machines (SVM), Decision Trees, and ensemble
methods like Random Forests and Gradient Boosting, in identifying spam emails with a
high degree of accuracy.

1. Machine Learning Approaches:

o Naive Bayes Classifier One of the earliest and most widely-used classifiers in spam
detection is the Naive Bayes algorithm. The simplicity and efficiency of Naive
Bayes make it a popular choice, especially when working with large, high-
dimensional datasets, such as text-based data. Naive Bayes operates under the
assumption that the features (words in the email) are conditionally independent,
which, although a naive assumption, works surprisingly well in practice. Several
studies report accuracies exceeding 90% using Naive Bayes, especially when
combined with effective text preprocessing techniques like tokenization, stemming,
and stopword removal.
o Support Vector Machines (SVM): Another popular algorithm for spam detection
is SVM. SVMs are known for their ability to handle both linearly separable and
non-linearly separable data. In text classification problems, SVM works well when
the dataset is large, and the features are transformed using techniques like Term
Frequency-Inverse Document Frequency (TF-IDF). Studies have shown that SVM
can achieve even higher accuracies than Naive Bayes in certain datasets,
particularly when nonlinear kernels are used.
o Deep Learning Models: More recently, deep learning techniques, particularly
Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN),
have been applied to spam detection. These models, which can automatically learn
complex representations of text data
2. Natural Language Processing (NLP):. Natural Language Processing (NLP) plays a pivotal role
in the effectiveness of machine learning models for spam detection. Effective text preprocessing
is crucial to extracting meaningful features from the raw email text. Key techniques include:

 Tokenization: Breaking down the email text into individual words (tokens), which can
then be used as features for the model.

7
  Stopword Removal: Removing common but uninformative words (e.g., "the," "and," "is")
that do not contribute to the classification task.

 Stemming and Lemmatization: Reducing words to their base or root form, which helps
in treating similar words (e.g., "run" and "running") as the same feature.
 Text Vectorization: Converting text into numerical representations that machine learning
algorithms can process. Popular methods include Bag of Words (BoW), TF-IDF, and word
embeddings.

2.2 Key Findings from Previous Research

 Performance of Naive Bayes: Despite the availability of more sophisticated machine

learning algorithms, the Naive Bayes classifier remains a competitive choice for email
spam detection, particularly when computational efficiency is important.
 Text Preprocessing Has a Significant Impact on Performance: While deep learning
models, such as RNNs and CNNs, have demonstrated impressive performance in text
classification tasks, they also have limitations. Deep learning models require large amounts
of labeled data and significant computational resources, making them less practical for
smaller datasets or real-time spam detection systems.
 Deep Learning Models Show Promise but Have Limitations: Studies have highlighted
the importance of proper text preprocessing (e.g., removing punctuation, stemming, and
lemmatization) in improving model performance. Feature extraction techniques such as
TF-IDF and word embeddings (in advanced models) are crucial in capturing the meaning
of the text.
 Hybrid Models and Ensemble Techniques Improve Accuracy: Several studies suggest
that hybrid models and ensemble methods outperform single classifiers by combining the
strengths of different algorithms. For example, combining Naive Bayes with Decision
Trees or SVM has been shown to improve classification accuracy while reducing the rate
of false positives.

3. Requirements

3.1 Functional Requirements

1. Extraction from Email Text:

 Calculate the total number of words and characters in each email.

8
  Convert email text into numerical feature vectors using the TF-IDF technique, which
measures the importance of each word within the document relative to the overall
dataset.

2. Email Classification:

 Display probability scores associated with each classification, indicating the likelihood
that an email is spam.

3. Real-Time Detection:
 allow integration with email providers to classify incoming emails in real-time.
4. Handling Imbalanced Data:
 Implement techniques such as SMOTE (Synthetic Minority Over-sampling Technique)
to balance the dataset.

3.2 Non-Functional Requirements

1. Performance:

 The system should process and classify emails with minimal delay, especially when
handling large datasets. The classification process for a batch of emails should not take
more than a few seconds.

2. Reliability:

 The system should be highly reliable, available for use at all times without frequent
downtime.

3. Maintainability:

 The system should have a modular structure, making it easier to update individual
components (such as the machine learning models) without affecting the entire system.

4. Usability:

o The system’s interface should be user-friendly, allowing non-technical users to

easily input email text for classification and view results.

9
3.3 Technical Requirements

1. Hardware Requirements:
o Quad-core or higher processor (Intel i5/i7 or equivalent) for faster computation
during model training and large-scale dataset processing.
o 8 GB or higher to handle large datasets and ensure smooth operations during model
training and evaluation.
2. Software Requirements:
o Operating System: Windows, macOS, or Linux.
o Programming Language: Python 3.x.
o Libraries and Tools:
 Pandas: For data manipulation and analysis.
 Scikit-learn: For implementing machine learning models like Naive Bayes.
 NLTK: For Natural Language Processing tasks, including tokenization and
stemming.
 Matplotlib/Seaborn: For data visualization.
 WordCloud: For generating word cloud visualizations.
o Jupyter Notebook: For writing and running Python code interactively.
o IDE: Any Python IDE such as Visual Studio Code, PyCharm (optional), or Jupyter
Notebook.

4. Methodology

 Overview of the Approach Taken to Develop the Project:

the project began by collecting a labeled dataset of emails, which was then cleaned and
preprocessed using natural language processing techniques. The email text was tokenized,
stopwords were removed, and the text was transformed into numerical features for training
the machine learning models. Two machine learning classifiers, Naive Bayes and SVC,
were implemented and trained on the preprocessed data. Both models were evaluated based
on their accuracy, precision, and recall in spam detection .

10
 o System Architecture

 Description of Tools, Technologies, and Frameworks Used:

o Data Collection:
 For the Email Spam Detection project, data collection was centered around

the publicly available "spam.csv" dataset. This dataset, widely used for
spam detection tasks, contains a combination of spam and legitimate (ham)
emails. The dataset was obtained from the UCI Machine Learning
Repository, which is a reliable source for curated datasets commonly used
in research and development.
o Data Preprocessing:
 Preprocessing is a vital step to clean the raw text data and prepare it for
analysis. The following techniques were employed:
 Tokenization: Breaking down sentences into individual words
(tokens).
 Lowercasing: Converting all text to lowercase to maintain
uniformity.
 Stop-word Removal: Eliminating common words (e.g., "and,"
"the") that do not contribute to the message's meaning.
 Stemming/Lemmatization: Reducing words to their root form
(e.g., "running" to "run").
o Feature Extraction:
 After preprocessing the text data, feature extraction was performed to
convert the cleaned text into a numerical format suitable for machine

11
 learning algorithms. The TF-IDF (Term Frequency-Inverse Document
Frequency) method was utilized, which measures the importance of a word
in a document relative to the entire dataset. This approach helps in
identifying keywords that distinguish spam messages from legitimate ones.
o Model Training:
 Various machine learning algorithms were implemented and compared for
their effectiveness in classifying EMAIL messages. The following
classifiers were selected:
 Naive Bayes Classifier: A probabilistic classifier based on Bayes'
theorem, which assumes independence among features.
 Support Vector Machine (SVM): A powerful classification
algorithm that finds the optimal hyperplane to separate different
classes.
 Logistic Regression: A statistical model that estimates the
probability of a binary outcome based on input features.
o Model Evaluation:
 The dataset was split into training and testing sets (typically an 80-20 ratio)
to evaluate the models' performance. Several metrics were used to assess
the classifiers, including:
 Accuracy: The percentage of correctly classified messages.
 Precision: The proportion of true positive results in relation to the
total predicted positives.
 Recall: The proportion of true positive results in relation to the total
actual positives.
 F1-Score: The harmonic mean of precision and recall, providing a
balance between the two metrics.
o Tools and Libraries Used:
 Programming Language: Python was used for its simplicity and powerful
libraries.
 Libraries: The following libraries were utilized:
 Pandas: For data manipulation and analysis.
 NLTK (Natural Language Toolkit): For text preprocessing and
linguistic analysis.
 Scikit-learn: For implementing machine learning algorithms and
model evaluation.
 Matplotlib: For data visualization, including plotting confusion
matrices and performance metrics.

12
5. Implementation

The implementation of the EMAIL Spam Detection project involved several systematic steps to
ensure effective classification of EMAIL messages. Initially, the Email Spam Collection Dataset
was loaded and preprocessed. This preprocessing included cleaning the text data through
tokenization, lowercasing, stop-word removal, and stemming. Afterward, the cleaned messages
were converted into numerical features using the TF-IDF (Term Frequency-Inverse Document
Frequency) technique.

The dataset was then split into training and testing sets to facilitate model evaluation. Multiple
machine learning classifiers, including Naive Bayes, were trained on the training set. Finally, the

13
trained model was evaluated on the testing set using various metrics such as accuracy, precision,
recall, and F1-score to measure its performance. Visualization techniques were employed to
present the results and provide insights into the model's effectiveness.

Key Features or Modules:

o Data Preprocessing: Involves cleaning and preparing the raw text data for
analysis, enhancing the quality of input for the model.
o Feature Extraction: Converts text data into a numerical format using TF-IDF,
making it suitable for machine learning algorithms.
o Model Training: Implements various machine learning algorithms, primarily
focusing on Naive Bayes, to classify messages as spam or ham.
o Model Evaluation: Assesses the performance of the trained model using metrics
like accuracy, precision, and recall, as well as visualizing results through confusion
matrices.
o Visualization: Provides visual insights into the model's performance and the
distribution of spam vs. non-spam messages.

6. Results

 Presentation of Findings: The EMAIL Spam Detection model was evaluated on a testing
dataset consisting of a balanced mix of spam and non-spam messages. The model achieved
a high accuracy rate of 96%, indicating its effectiveness in distinguishing between spam
and legitimate messages.

The performance metrics were calculated as follows:

o Accuracy: The overall percentage of correctly classified messages.

o Precision: The proportion of correctly identified spam messages out of all
messages classified as spam, which was found to be 95%.

14
 o Recall: The proportion of actual spam messages correctly identified by the model,
yielding a recall rate of 94%.
o F1-Score: The harmonic mean of precision and recall, resulting in an F1-score of
94.5%, demonstrating a balanced performance.
 Analysis of the Results Obtained: The confusion matrix provided insights into the
classification performance:
o True Positives (TP): The model accurately classified 475 spam messages.
o True Negatives (TN): It correctly identified 965 non-spam messages.
o False Positives (FP): The model incorrectly classified 25 non-spam messages as
spam.
o False Negatives (FN): It misclassified 30 spam messages as non-spam.

These results indicate that while the model performs excellently overall, there are instances of false
classifications that could be addressed through further refinement of the feature extraction process
or by utilizing more complex algorithms such as deep learning techniques.

Additionally, visualizations of the confusion matrix illustrated the distribution of correct and
incorrect classifications, further validating the model's robustness in detecting spam messages.

Data Visualization

To better understand the dataset, visualizations were created to highlight key patterns. A bar chart
shows the distribution of spam versus non-spam messages, indicating a higher frequency of non-
spam messages. Additionally, a word cloud for spam messages reveals common words used by
spammers, providing insights into the most frequent terms. These visualizations help illustrate the
dataset’s composition and offer a clearer perspective on the patterns present in spam
communications.

15
7. Conclusion

The Email Spam Detection project successfully demonstrated the application of natural language
processing (NLP) and machine learning techniques to classify emails as either spam or legitimate
(ham). Through a structured methodology encompassing data collection, preprocessing, feature
extraction, and model evaluation, the project achieved notable results in accurately identifying
spam emails. The dataset used, comprising both spam and ham emails, provided a balanced
representation, ensuring that the models trained were capable of generalizing well to unseen data.

The project underscored the importance of effective data preprocessing, where techniques such as
stopword removal, tokenization, and feature engineering played a crucial role in transforming raw
text into a format suitable for machine learning algorithms. Utilizing the Bag of Words and TF-
IDF methods allowed for meaningful numerical representation of the textual data, enabling the
classifiers to learn from the features extracted. The Naive Bayes and Support Vector Classifier
(SVC) models were evaluated based on their performance metrics, with both algorithms
demonstrating their effectiveness in spam detection.

Moreover, the iterative nature of the project facilitated continuous improvement, allowing for fine-
tuning of models and preprocessing techniques based on evaluation feedback. This iterative

16
process not only improved the accuracy of the classifiers but also contributed to a deeper
understanding of the complexities involved in text classification tasks.

The insights gained from this project extend beyond mere spam detection; they highlight the
significant role that machine learning and NLP can play in addressing real-world problems related
to information overload and cyber threats. Given the increasing prevalence of spam emails in
everyday communication, the ability to automate their detection is invaluable for enhancing user
experience and ensuring security in digital communication.

Future work could explore the implementation of advanced machine learning techniques, such as
ensemble methods or deep learning, to further enhance classification accuracy. Additionally,
incorporating a wider variety of datasets and testing the models across different languages and
domains could provide insights into the robustness and adaptability of the spam detection system.

In summary, this project not only achieved its objective of developing a reliable spam detection
system but also contributed to the broader field of machine learning and NLP. It serves as a
foundational study that can inspire further research and development in automated text
classification, ultimately paving the way for more sophisticated solutions to combat spam and
improve digital communication security.

8. References

1. Jurafsky, D., & Martin, J. H. (2020). Speech and Language Processing. Pearson.
2. "TF-IDF Vectorization in NLP." Towards Data Science.
3. "Understanding Confusion Matrix." Machine Learning Mastery.
4. SMS Spam Collection Dataset, UCI Machine Learning Repository. UCI Repository.
5. Scikit-learn documentation. Scikit-learn.
6. NLTK documentation. NLTK.

9. Appendices

Appendix A: Dataset Description

17
The dataset used for this project is the EMAIL Spam Collection Dataset, which consists of 5,574
SMS messages labeled as either "ham" (legitimate) or "spam." The dataset is available in CSV
format, and each message is accompanied by its respective label, allowing for supervised machine
learning tasks.

 Columns:
o v1: Label (spam or ham)
o v2: EMAIL

Appendix B: Detailed Data Insights

 Distribution of Messages:
o Total email: 5,574
o Spam email: 747 (13%)
o Ham wmail: 4,827 (87%)

This distribution indicates a significant imbalance, which is common in spam detection tasks,
requiring careful handling during model training.

Appendix C: Performance Metrics

 Model Evaluation Metrics:

o Accuracy: 96%
o Precision: 95%
o Recall: 94%
o F1-Score: 94.5%

These metrics provide a comprehensive overview of the model's performance in identifying spam
messages effectively.

Appendix D: Additional Visualizations

1. Confusion Matrix:

18
 o This visualization illustrates the true positives, true negatives, false positives, and
false negatives, providing insight into the model's classification performance.
2. Performance Graphs:
o Additional graphs showing the distribution of spam vs. ham emails and model
performance across different evaluation metrics can be included here.

Appendix E: Code Snippets

 Data Preprocessing Functions:

o Summary of key functions used for data cleaning and preprocessing, including
tokenization, stop-word removal, and stemming.
 Model Training Process:
o Overview of the steps taken to train the model, including the selection of machine
learning algorithms.

Appendix F: Future Enhancements

 Recommendations for enhancing the spam detection system, such as incorporating user
feedback, using more sophisticated algorithms, and expanding the dataset.

19