README.md

metadata

title: Milk Spoilage Classifier - Multi-Variant API
emoji: 🥛
colorFrom: indigo
colorTo: purple
sdk: docker
pinned: false
license: mit

Milk Spoilage Classification API - Multi-Variant

AI-powered milk spoilage classification with 10 specialized model variants optimized for different data availability scenarios.

🚀 Features

• 10 Model Variants with test accuracies from 62.8% to 95.8%
• Automatic Feature Validation - API validates required features for each variant
• RESTful API with comprehensive OpenAPI documentation
• Custom GPT Ready - Designed for seamless ChatGPT integration

📊 Available Model Variants

Rank	Variant	Test Accuracy	Features Required
🥇	baseline	95.8%	All 6 features
🥈	scenario_1_days14_21	94.2%	Days 14 & 21 (SPC+TGN)
🥉	scenario_3_day21	93.7%	Day 21 only (SPC+TGN)
4	scenario_4_day14	87.4%	Day 14 only (SPC+TGN)
5	scenario_2_days7_14	87.3%	Days 7 & 14 (SPC+TGN)
6	scenario_6_spc_all	78.3%	SPC only (All Days)
7	scenario_8_spc_7_14	73.3%	SPC only (Days 7 & 14)
8	scenario_9_tgn_7_14	73.1%	TGN only (Days 7 & 14)
9	scenario_7_tgn_all	69.9%	TGN only (All Days)
10	scenario_5_day7	62.8%	Day 7 only (SPC+TGN)

🔧 API Endpoints

GET /variants

List all available model variants with metadata

POST /predict

Make a prediction using the specified model variant

Example Request:

{
  "spc_d7": 2.1,
  "spc_d14": 4.7,
  "spc_d21": 6.4,
  "tgn_d7": 1.0,
  "tgn_d14": 3.7,
  "tgn_d21": 5.3,
  "model_variant": "baseline"
}

Example Response:

{
  "prediction": "PPC",
  "probabilities": {
    "PPC": 0.97,
    "no spoilage": 0.02,
    "spore spoilage": 0.01
  },
  "confidence": 0.97,
  "variant_used": {
    "variant_id": "baseline",
    "name": "Baseline (All Features)",
    "test_accuracy": 0.9576,
    "features": ["SPC_D7", "SPC_D14", "SPC_D21", "TGN_D7", "TGN_D14", "TGN_D21"]
  }
}

📖 Spoilage Classes

Class	Description
PPC	Post-Pasteurization Contamination - Bacteria introduced after pasteurization
no spoilage	No significant spoilage detected
spore spoilage	Heat-resistant spore-forming bacteria survived pasteurization

🔗 Interactive Documentation

Visit /docs for interactive Swagger UI documentation where you can test the API directly.

💻 Usage Example

curl -X POST https://chenhaoq87-milkspoilageclassifier-api-variants.hf.space/predict \
  -H "Content-Type: application/json" \
  -d '{
    "spc_d21": 6.4,
    "tgn_d21": 5.3,
    "model_variant": "scenario_3_day21"
  }'

🤖 Custom GPT Integration

This API is designed for Custom GPT integration. The GPT will automatically select the best variant based on available data.

See the Custom GPT Setup Guide for complete integration instructions.

📝 Input Format

All microbial count values should be in log CFU/mL (base 10):

• SPC (Standard Plate Count): Total bacterial count
• TGN (Total Gram-Negative): Gram-negative bacteria count
• Measured at Day 7, 14, and 21

🔍 Variant Selection Guide

• Have all measurements? → Use baseline (best accuracy)
• Only Day 21 data? → Use scenario_3_day21 (nearly as good!)
• Only Day 14 data? → Use scenario_4_day14
• Only SPC measurements? → Use scenario_6_spc_all
• Only TGN measurements? → Use scenario_7_tgn_all

📚 Research

Based on predictive modeling for milk spoilage classification using microbial growth patterns.

---

Model Repository: MilkSpoilageClassifier