🔐 Online Security System

This project predicts whether a website is safe or phishing/malicious, using a set of well-known features commonly used in phishing detection systems. It is a classification problem built with a robust data pipeline, feature validation, and multiple machine learning models.

---

📊 Problem Statement

Cybersecurity threats like phishing attacks are increasingly prevalent. This system leverages categorical features derived from URLs and website behavior to classify websites as safe or unsafe.

The dataset used in this project is sourced from Kaggle and contains a variety of features such as IP address usage, URL length, HTTPS presence, domain registration, traffic ranking, and more.

---

🚀 Features Used

Feature Name	Description
having_IP_Address	Uses IP instead of domain name
URL_Length	Length of the full URL
Shortining_Service	Uses URL shorteners like bit.ly
having_At_Symbol	Presence of "@" in the URL
double_slash_redirecting	Use of // for redirection
Prefix_Suffix	Hyphen in domain name
having_Sub_Domain	Number of subdomains
SSLfinal_State	Validity of SSL certificate
Domain_registeration_length	Length of domain registration
Favicon	Favicon loaded from external source
port	Use of non-standard ports
HTTPS_token	Misleading "https" in domain
Request_URL	External objects in URL
URL_of_Anchor	Anchor tags leading to other domains
Links_in_tags	Meta/script/link tag links
SFH	Form handler behavior
Submitting_to_email	Form submits to email address
Abnormal_URL	URL differs from page source
Redirect	Number of redirections
on_mouseover	Changes on mouse hover
RightClick	Right click disabled
popUpWidnow	Presence of pop-ups
Iframe	Use of invisible iframes
age_of_domain	Age of the domain
DNSRecord	Availability of DNS info
web_traffic	Site traffic rank
Page_Rank	Google's page ranking
Google_Index	Indexed by Google or not
Links_pointing_to_page	Inbound links to the page
Statistical_report	Reported in phishing/malware lists

---

⚙️ Tech Stack

Language: Python 3.8+

Backend:

• FastAPI
• Uvicorn (ASGI Server)

Machine Learning:

• Scikit-learn
• NumPy
• Pandas

Experiment Tracking & Serialization:

• MLflow
• dill

Database:

• MongoDB Atlas (Cloud NoSQL)
• PyMongo

Configuration & Environment:

• python-dotenv
• certifi
• pyaml

---

🧱 Project Architecture

MongoDB Atlas: Raw data is pushed to a NoSQL database hosted on the cloud.

Data Ingestion: Loads data from MongoDB and splits it into training and test sets.

Data Validation: Validates schema, checks for missing values, and generates data drift reports.

Data Transformation: Prepares the dataset for training by encoding features.

Model Training: Trains and evaluates models on the dataset.

Artifact Tracking: Each component generates and logs artifacts (train/test data, models, metrics).

---

🧪 Models Used

The following models are trained and compared based on classification metrics:

'Random Forest': RandomForestClassifier(verbose = 1)

'Decision Tree': DecisionTreeClassifier()

'Gradient Boosting': GradientBoostingClassifier()

'Logistic Regression': LogisticRegression(verbose = 1)

'AdaBoost': AdaBoostClassifier()

---

📦 Installation

1. Clone the repository- git clone https://github.com/your-username/online-security-system.git cd online-security-system

2. Create and activate a virtual environment (optional but recommended)

3. Install the required dependencies- pip install -r requirements.txt

4. Environment Setup- MONGODB_URL="your_mongodb_atlas_connection_string"

5. run in terminal- uvicorn app:app --reload

This will open the interactive FastAPI Swagger UI, where you can test the API endpoints for training and prediction.

---