🔍 Unsupervised Anomaly Detection on Tabular Data

This project is a general-purpose anomaly detection tool using unsupervised machine learning, designed to work on any CSV dataset. Whether it’s detecting fraud in financial transactions or identifying abnormal behavior in refinery equipment, this tool lets you clean data, apply Isolation Forest, and detect anomalies — all without needing labeled data.

✅ Bonus: Includes a Streamlit GUI that allows users to upload their own datasets and detect anomalies interactively.

---

🧠 Project Overview

• 📌 Goal: Automatically detect anomalies (outliers) in tabular data where labels like “Normal” or “Faulty” are not available
• 🧠 ML Algorithm: Isolation Forest (Unsupervised Learning)
• 🛠️ Use Cases:
  • Industrial asset monitoring (refinery sensor logs)
  • Credit card fraud detection
  • Network/server failure prediction
  • IoT device anomaly tracking
• ⚙️ Tech Stack: Python, Pandas, Scikit-learn, Seaborn, Streamlit, Jupyter Notebook

---

🗂️ Folder Structure

unsupervised-anomaly-detector/
├── streamlit_app.py
├── requirements.txt # Dependencies
└── README.md # Project documentation

---

💡 How It Works

1. User Input: Upload any tabular .csv dataset (e.g., equipment logs, transaction records, sensor data).
2. Preprocessing: Cleans missing values, scales numerical data.
3. Anomaly Detection: Applies Isolation Forest to detect anomalies based on data behavior.
4. Output: Labels rows as "Normal" or "Anomaly" and optionally exports the result.

---

📦 Technologies Used

Tool	Role
Python	Core programming language
Pandas	Data manipulation
Scikit-learn	ML algorithm (Isolation Forest)
Seaborn	Visualizations
Streamlit	GUI for interactive anomaly detection
Jupyter	Prototyping and step-by-step learning

---

🌐 Streamlit Web App

This app allows non-technical users to upload any CSV file and run anomaly detection interactively.

🔧 Features:

• Upload any .csv file
• Auto-cleaning of missing/null values
• Scaling of numeric features
• Anomaly detection using Isolation Forest
• Visual summary of results
• Downloadable output with anomaly labels

---

🚀 To Launch the App:

cd streamlit_app
streamlit run app.py

⚙️ Setup Instructions

1. Clone the Repository

git clone https://github.com/ArshSharan/Unsupervised-Anomaly-Detection.git
cd unsupervised-anomaly-detector

2. Install Dependencies

pip install -r requirements.txt

3. Run the Notebook

jupyter notebook
# Open `Unsupervised_Detection.ipynb`

4. Launch Streamlit App

cd streamlit_app
streamlit run app.py

🔧 Real-World Use Cases

Domain	Example Use
🔧 Oil & Gas	Detect abnormal equipment behavior using sensor logs
💳 Finance	Spot fraudulent transactions
📶 IoT Devices	Monitor for unexpected spikes
🏭 Manufacturing	Detect process anomalies

🔍 Future Additions

• Add One-Class SVM and Autoencoders for comparison

• Feature importance and SHAP visualizations

• Live data ingestion with MQTT or REST APIs

• Deployment on cloud (Heroku, Streamlit Cloud, etc.)

🙋‍♂️ Who Is This For?

• Data science learners.

• ML engineers building tools.

• Interns working on anomaly detection projects.

⭐ Like This Project?

Give it a ⭐ if it helped you or inspired you to build something cool!

---