KALLAM HARANADHAREDDY INSTITUTE OF TECHNOLOGY

(AUTONOMOUS)
Approved by AICTE- New Delhi, Accredited by NAAC with ‘A’
Grade
Permanently Affiliated to JNTUK, Kakinada
NH-5, Chowdavaram, Guntur
DEPARTMENT OF CSE(ARTIFICIAL INTELLIGENCE & DATA
SCIENCE)
SMS SPAM DETECTION

SUBMITTED BY
1.SD.SAJIDA
THARUNAM(218X1A4539)
2.P.RUKMINI LAKSHMI
CHAITANYA(218X1A4528) GUIDE NAME:
3. R.ESHWAR REDDY(218X1A4557) Mr. K.SRINIVASA
4.CH.SATYA PRANITH(218X1A4556) REDDY
SMS
Spam
Detecti
on
 INDEX
 Abstract  Proposed method

 Objective of project  Advantages

 Problem Statement  Project Flow

 Scope & Motivation  Hardware and Software

Requirements
 Introduction
 Architecture
 Literature survey
 Module
 Existing Method
 Output Screens
 Disadvantages
 Conclusion

 Future Work References

 Abstra
ct
The SMS Spam Detection project tackles the increasing problem of
unsolicited and potentially harmful SMS messages by utilizing
machine learning methods. Spam messages often consist of
unwanted
advertisements, phishing attacks, and fraud schemes that
threaten user security and reduce productivity. The goal of this
project is to develop a Flask-based web application that detects
and classifies
incoming SMS messages as Spam or Not Spam in real-time. The
heart of the project is a Naive Bayes classifier trained on a
dataset of over 5,000 SMS messages sourced from platforms like
Kaggle. By
incorporating advanced text processing techniques such as TF-IDF
vectorization, the project ensures accurate predictions, offering
users a scalable, user-friendly solution for spam filtering.
 Our project seeks to mitigate this

0 problem by leveraging machine

learning to build an automated
system that classifies SMS messages

1.
as either spam or not spam.

A Naive Bayes classifier was selected

Key
for this project due to its effectiveness

0 in text classification, combined with

TF-IDF (Term Frequency-Inverse
Document Frequency) vectorization to

Objectives 2.
handle the text data accurately.

We designed a Flask-based web

0 application that allows real-time

spam classification, offering users
a convenient, accessible solution.

3.
 Objective of the
Project
Accurate Spam Detection: Develop a robust
model to classify SMS messages with high accuracy,
minimizing false positives and negatives.

User-Friendly Interface: Create an intuitive web

application where users can enter SMS messages
and instantly receive a classification result.

Scalable Real-Time Performance: Ensure the

system operates with low latency, handling
classification requests in real-time, and maintaining
performance when scaled.
 Problem Statement
Security Threats: Spam messages are not only intrusive but also present significant security
risks. Many of these messages contain phishing links or deceptive content designed to steal
personal information or manipulate users into fraudulent activities, putting data privacy and user
safety at risk.

Productivity Impact: Unsolicited messages flood users' inboxes, causing distractions and
making it difficult to locate important messages. This not only wastes time but also impacts
productivity as users have to manually sift through and delete these unwanted messages.

Economic and Social Consequences: Spam can lead to financial losses for individuals and
organizations through scams or compromised security. It also erodes trust in messaging platforms,
as users become more cautious and may ignore legitimate messages.

Limitations of Traditional Filtering: Conventional filtering methods, such as keyword-based

approaches, are limited in their effectiveness. Spammers frequently update their tactics by using
obfuscated text, non-standard language, or changing message patterns, rendering traditional
filters inadequate.

Need for an Intelligent Solution: An adaptive, machine learning-based solution is necessary to

effectively combat the evolving nature of SMS spam. Such a system can dynamically learn from
new spam patterns, providing more robust and real-time protection for users.
 SCOPE & MOTIVATION

Scope: Motivation:

Scalability: This project can be scaled The rise in SMS spam has heightened
for different platforms, including mobile the need for a secure, automated
apps and integration with SMS services. filtering solution to protect user
privacy, avoid data loss, and increase
Multi-Language Support: The model productivity.
can be expanded to classify messages
in multiple languages with additional By leveraging machine learning, we
training data. aim to provide an efficient and
scalable solution to address this
Adaptability: The system can be pervasive issue.
regularly updated to adapt to evolving
spam tactics.
 INTRODUCTION
•Prevalence of SMS Spam: The rise of SMS as a popular communication channel has also led to a
surge in spam messages, which often include unsolicited ads, fraudulent offers, and phishing links
aimed at deceiving users. This pervasive issue affects millions globally, leading to privacy
breaches, financial losses, and user frustration.

•Challenges in Detecting Spam: Traditional filtering methods are limited in scope, often failing to
keep up with the rapidly evolving tactics used by spammers. Simple keyword-based filters can be
easily bypassed, making it essential to adopt more advanced, intelligent solutions.

•The Role of Machine Learning: Machine learning provides a powerful approach for detecting
spam by analyzing patterns and recognizing nuanced differences between spam and legitimate
messages. Models like Naive Bayes can process large volumes of text data, learning from new
types of spam and adapting to variations in language and structure.

•Project Objective: This project leverages machine learning techniques to develop an accurate,
real-time SMS spam detection system. By incorporating advanced text processing and classification
methods, the system aims to deliver a scalable, user-friendly tool that provides effective spam
filtering for users.

•Project Aim: This project harnesses these capabilities to deliver a real-time, efficient, and user-
friendly spam detection system.
 Literature Survey
Overview of Existing Research:

Bag of Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF) are
commonly used techniques for converting text data into feature vectors for machine learning models.
•Algorithms such as Naive Bayes, SVM, and LSTM have been used for spam detection, with Naive
Bayes being popular due to its fast and simple implementation.

Research Paper 1:

•"SMS Spam Detection using Naive Bayes Classifier" by Khin Sandar Win et al. focuses on
the use of Naive Bayes for text classification in SMS spam detection. It demonstrates the algorithm’s
efficiency in handling large datasets and producing high accuracy 【 250†source 】 .

Research Gap:

•Many studies focus on achieving high accuracy in detecting spam but fall short in minimizing false
positives.
•Few projects have focused on real-time deployment, which is essential for applications like SMS spam
detection in everyday mobile or web use.
 Existing Methods
Traditional Methods: Machine Learning-Based Systems:

Keyword-based Filtering: Naive Bayes:

• Relies on predefined spam keywords or • A probabilistic classifier that is particularly
patterns. However, spammers can easily effective for text classification problems like
modify message content to bypass filters. spam filtering. It assumes independence
between features and is computationally
Blacklists: efficient.
• Blocks messages from known spam
sources. It is static and cannot handle new Support Vector Machines (SVM):
or unknown spam sources. • Effective for text classification but
computationally expensive due to the need for
Rule-based Filtering: hyperparameter tuning and non-linear kernel
• Uses hard-coded rules to classify calculations.
messages. Rules must be updated
manually, and the system is inflexible. Deep Learning Approaches:
• Models like LSTMs and RNNs can capture
sequential patterns in text but require significant
computational power and large datasets for
training.
 Disadvantages of Existing
Methods
Keyword-Based Filtering:

• Low Accuracy: Simple alterations to message content can bypass keyword filters, leading to low
detection rates.

Rule-Based Systems:

• High False Positives: Legitimate messages may be flagged as spam if they match predefined rules
too strictly.

•Machine Learning-Based Systems:

• Data Dependency: Requires a large, labeled dataset for training, and insufficient training data can
result in poor performance.

• Complexity: Advanced models like SVM and LSTMs are computationally expensive and can be
difficult to interpret (black-box models).

• Deployment Challenges: Real-time applications demand low latency and scalability, which can be
difficult to achieve with resource-heavy models like SVM.
 Proposed Method
Methodology:

Step 1: Data Preprocessing:

• Clean the SMS messages (removing special characters, URLs,
etc.).
• Convert the text into numerical features using
CountVectorizer or TF-IDF.

Step 2: Model Training:

• Use Naive Bayes classifier, which is suitable for large-scale
text classification tasks. The model is trained on labeled SMS
data (spam vs. non-spam).

Step 3: Model Evaluation:

• Evaluate the model’s performance using metrics like accuracy,
precision, recall, and F1-score.

Step 4: Deployment:
• The trained model is deployed using the Flask framework to
provide real-time spam detection via a web interface.
Advantages
Efficiency: The Naive Bayes algorithm is lightweight and
computationally efficient, making it well-suited for text classification
tasks and enabling rapid processing even on limited hardware.

Real-Time Detection: With the Flask-based web application, users

receive instant spam classification results, offering a seamless and
responsive experience.

Scalability: Built with scalability in mind, the system can be deployed

on cloud infrastructure to handle a growing user base and high
message volumes without compromising performance.

Adaptability: The machine learning model can be periodically updated

with new data, allowing it to learn from evolving spam patterns and
stay effective against new threats.

User Accessibility: Designed as a web application, this solution is

accessible to users with varying technical backgrounds, providing an
intuitive and hassle-free way to detect spam.
 PROJECT FLOW
 User Inputs SMS Text: The user enters the SMS message into the input field on the web
interface, initiating the spam detection process.

 Text Preprocessing: The input message undergoes preprocessing, which includes:

• Cleaning: Removing special characters, links, and unnecessary whitespace.
• Tokenization: Breaking the text into individual words (tokens) to facilitate analysis.
• Vectorization: Using TF-IDF to convert the cleaned and tokenized text into a numerical
format that the model can process.

 Model Prediction: The preprocessed text is fed into the Naive Bayes classifier, which
calculates the likelihood of the message being spam based on learned patterns.

 Result Display: The classification result—either "Spam" or "Not Spam"—is displayed on the
web interface, providing immediate feedback to the user.

 Feedback and Logging (Optional): The system can log results or accept user feedback to
further improve model accuracy over time, making the solution more adaptable to new spam
patterns.
 Hardware and Software
Requirements
Hardware Requirements:

 CPU: 2 GHz processor or higher

 Memory: Minimum 4GB RAM

Software Requirements:

 Programming Language: Python

 Libraries: Scikit-Learn for machine learning, NLTK for text processing, Flask for web app
development

 Web Development: HTML, CSS, Bootstrap for frontend design

 Deployment Options: Compatible with cloud platforms (e.g., AWS, Azure) for scalability.
ARCHITECTURE
 MODULE
1. Data Preprocessing Module:

 Cleans SMS data by removing unwanted characters, duplicates, and normalizing text.
 Example: A message "Congratulations! You've won a FREE iPhone!" is converted to "congratulations
you’ve won a free iphone".

2. Model Training Module:

 Trains the selected machine learning algorithms (Naive Bayes, Random Forest, SVM) on the cleaned
dataset.
 Example: Splitting the dataset into training (80%) and testing (20%) sets for model evaluation.

3. Prediction Module:

 Takes new incoming SMS messages and classifies them using the trained model.
 Example: Classifying "Get paid to work from home!" as spam based on learned patterns.

4. User Interface Module:

 A simple web interface to display the classification result to users.

 Example: If a message is classified as spam, an alert is displayed: "Warning: This message may be
spam."
 FUNCTIONALITIES
Home Page:

• Description: The home page provides a user-friendly interface with a clear title, "SMS Spam Detection". It includes a button
labeled "Click Here for Spam Detection". When the user clicks this button, they are redirected to the main spam detection
module.

Upload SMS:

• Description: This module allows users to input or paste SMS messages for analysis. The text of the SMS is submitted through
a web form, which is processed by the system to detect whether the message is spam or not.

Spam Detection:

• Description: After the SMS is submitted, the spam detection module processes the message using a pre-trained machine
learning model (e.g., Naive Bayes classifier). The system analyzes the content and classifies the message as either spam or not
spam based on learned patterns from the training data.

Spam Analysis:

• Description: This module provides detailed insights into why a message was classified as spam or not spam. It highlights key
factors such as suspicious words, links, or patterns that contributed to the spam classification, helping users understand the
reasoning behind the detection.

Result Display:

• Description: The result is displayed to the user, indicating whether the SMS is spam or not. The module may also display
additional information, such as a confidence score or keywords that triggered the classification, enhancing user transparency.
OUTPUT SCREENS
Conclusi
on
This project successfully addresses the pervasive issue of SMS spam by
leveraging machine learning and advanced text processing techniques.
Using the Naive Bayes algorithm combined with TF-IDF vectorization, we
built a model that achieves high accuracy in classifying messages as spam
or non-spam. This method not only improves spam detection but also
adapts to various types of messages, ensuring a robust filtering system.

The Flask web application offers a user-friendly interface, making the tool
accessible to a broad audience, including those without technical expertise.
Its real-time prediction capability provides users with immediate feedback,
enhancing the user experience and allowing for quick decision-making.

Additionally, the design of the system allows for future scalability and
adaptability. As spam techniques evolve, the model can be retrained on
updated data to maintain effectiveness. The application is also ready for
deployment on cloud platforms, enabling it to handle larger user volumes
and diverse usage scenarios.

In summary, this project provides an efficient, scalable, and adaptable

solution for SMS spam detection, contributing to enhanced user security,
productivity, and peace of mind.
 Future
.
Work
Model Improvement: Experiment with advanced deep
learning models such as LSTM or BERT for better context
understanding.

Continuous Learning: Implement mechanisms for the

model to learn from user feedback to improve
classification over time.

Broader Dataset: Collect and train on a more extensive

and diverse dataset to capture a wider variety of spam
types and languages.

Mobile Application: Develop a mobile app version of the

spam detection system for on-the-go protection.
THANK YOU!