Model Management System

The model management system allows dynamic loading and configuration of machine learning models for traffic classification.

Overview

The system provides:

• JSON-based model configuration
• Dynamic model loading/unloading
• Automatic category management
• Model validation and health checks
• API endpoints for management

Configuration File

Models are defined in classifier/models_config.json:

{
  "models": {
    "complex_cnn_16_04_2025": {
      "name": "Complex CNN (16-04-2025)",
      "model_path": "classifier/ml_models/complex_cnn_16-04-2025.h5",
      "model_type": "keras_h5",
      "num_categories": 22,
      "description": "Complex CNN model trained on 22 categories",
      "version": "16-04-2025",
      "confidence_threshold": 0.7,
      "is_active": true,
      "categories": [
        "ADS_Analytic_Track",
        "AmazonAWS",
        "BitTorrent",
        "Facebook",
        "GMail",
        "Google",
        "HTTP",
        "NetFlix",
        "TikTok",
        "YouTube"
      ]
    }
  }
}

Model Configuration Fields

Field	Type	Description
name	string	Human-readable model name
model_path	string	Path to model file
model_type	string	Model type (keras_h5)
num_categories	integer	Number of classification categories
description	string	Model description
version	string	Model version
confidence_threshold	float	Minimum confidence for classification
is_active	boolean	Whether model is currently active
categories	array	List of category names

API Endpoints

List Models

GET /api/v1/classifier/models/

Get Model Details

GET /api/v1/classifier/models/{id}/

Activate Model

POST /api/v1/classifier/models/{id}/activate/

Deactivate Model

POST /api/v1/classifier/models/{id}/deactivate/

Model Loading

Models are loaded dynamically when activated:

1. System reads configuration
2. Validates model file exists
3. Loads TensorFlow/Keras model
4. Verifies category count matches
5. Sets as active model

Category Management

Categories are automatically managed:

• Generated from model configuration
• Stored in database
• Used for classification results
• Displayed in UI

Performance Considerations

• Models are loaded into memory
• Only one active model at a time
• Model switching requires reload
• Consider memory usage for large models

Next Steps

• Advanced Configuration - System configuration