Viking Machine Learning Ensemble Model

This repository contains the backend for the document authenticity verification system. It employs an Ensemble Learning approach, combining predictions from six different machine learning models to achieve highly robust and reliable classification of certificates as "Genuine" or "Fraud."

The API accepts an image path, runs a simulated Computer Vision (CV) pipeline to extract key similarity scores, and feeds those scores to the ensemble for the final prediction.

---

Key Features

• 6-Model Ensemble: Combines the predictive power of CNN, Random Forest, SVM, Logistic Regression, K-Means, and Naive Bayes.
• Deep Learning Feature Integration: Uses simulated outputs from an image-based feature extractor (signature similarity, seal similarity, layout alignment).
• FastAPI Deployment: Robust and fast API server built on FastAPI for easy prediction serving.
• Tie-Breaker Priority: Uses a structured voting system with a clear priority (CNN > RF > SVM > LR > K-Means > NB) to resolve 3-3 ties.

---

Setup & Installation

1. Clone Repository

git clone [https://github.com/RishiGoswami-code/VikingMachineLearningModel.git]
cd VikingMachineLearningModel

---

2. Create & Activate Virtual Environment

python3 -m venv venv
source venv/bin/activate   # On Windows: .\venv\Scripts\activate

---

3. Install Dependencies

pip install fastapi uvicorn numpy scikit-learn tensorflow pydantic

---

Running the API Server

Start the Server

```bash
uvicorn app:app --reload --host 0.0.0.0 --port 8000
```

When the system loads and trains all models, you will see:

All 6 Models Trained and Ready.

Our API will run at:

http://127.0.0.1:8000

---

Swagger API Documentation

Open in your browser:

http://127.0.0.1:8000/docs

This interactive UI lets you test predictions directly.

---

API Usage

Prediction Endpoint

Method	Path	Description
POST	/predict	Runs the full ML pipeline + ensemble.

---

Input Format (JSON)

{
  "image_path": "certificate_good_sig_good_seal_high_clarity.jpg"
}

Output Example (JSON)

{
  "prediction_status": true,
  "prediction_label": "GENUINE"
}

---

Response Field Description

**Field **	**Type **	Description
prediction_status	boolean	true = Genuine, false = Fraud
prediction_label	string	"GENUINE" or "FRAUD"

---

Ensemble Voting Logic

The final decision is determined using majority voting among all six models:

• CNN

• Random Forest

• SVM

• Logistic Regression

• K-Means

• Naive Bayes

---

Tie-Breaker Priority (Highest → Lowest)

If the ensemble has a 3–3 split, the model with higher priority wins:

• CNN

• Random Forest

• SVM

• Logistic Regression

• K-Means

• Naive Bayes

---

How the System Works

image_analyzer.py simulates feature extraction from the provided image path.

Features are fed into all six ML models.

• Each model outputs its prediction.

• The ensemble computes majority vote.

• If tied, priority order decides the final label.

---

Technologies Used:

• FastAPI (API framework)

• TensorFlow (CNN model)

• scikit-learn (RF, SVM, Logistic Regression, Naive Bayes, K-Means)

• Uvicorn (ASGI server)

• Pydantic (input validation)

---

Future Enhancements

• Replace simulated CV pipeline with real image analysis

• Model retraining from UI

• Add confidence scores

• Deploy to cloud (AWS / GCP / Render)

Author:

Chandan GIri