---
language:
- en
license: apache-2.0
tags:
- text-classification
- blood-donation
- emergency-detection
- medical
- healthcare
- sri-lanka
- distilbert
- fine-tuned
datasets:
- AshenFdo/synthetic_blood_request_urgency_dataset
base_model: distilbert/distilbert-base-uncased
pipeline_tag: text-classification
metrics:
- accuracy
model-index:
- name: emergency_blood_request_classifier
  results:
  - task:
      type: text-classification
    dataset:
      name: synthetic_blood_request_urgency_dataset
      type: AshenFdo/synthetic_blood_request_urgency_dataset
    metrics:
    - type: accuracy
      value: 1.0
---

# 🩸 Emergency Blood Request Classifier

A fine-tuned **DistilBERT** model for **binary text classification** that automatically determines whether a blood donation request is an **emergency** or **not an emergency**.

This model is part of my personal project to build an AI-powered blood donation mobile application for Sri Lanka — designed to prioritize life-critical requests and alert nearby donors faster.

---

## 📌 Model Summary

| Property | Details |
|---|---|
| **Base Model** | [distilbert/distilbert-base-uncased](https://huggingface.co/distilbert/distilbert-base-uncased) |
| **Task** | Binary Text Classification |
| **Labels** | `emergency`, `not_emergency` |
| **Training Dataset** | [AshenFdo/synthetic_blood_request_urgency_dataset](https://huggingface.co/datasets/AshenFdo/synthetic_blood_request_urgency_dataset) |
| **Evaluation Accuracy** | **100%** (eval set) |
| **Training Epochs** | 10 |
| **Language** | English |
| **Domain** | Healthcare / Blood Donation (Sri Lanka) |
| **License** | Apache 2.0 |

---

## 🎯 Intended Use

### Primary Use
This model is designed to be integrated into a **Sri Lanka blood donation mobile application**. When a user posts a blood request, the model reads the description and classifies it as:

- 🚨 `emergency` → triggers immediate alerts to nearby donors
- 📋 `not_emergency` → listed normally in the donor feed

### How It Fits the Bigger Picture

```
Blood Donation App (Sri Lanka)
        │
        ▼
  User posts a blood request
        │
        ▼
  🤖 This Model (Emergency Classifier)
  ─────────────────────────────────────
  Reads the request description
        │
        ├──▶ emergency     → 🚨 Immediately alert nearby donors
        │
        └──▶ not_emergency → 📋 List normally in the donor feed
```

### Other Potential Uses
- Urgency triage in healthcare communication platforms
- Benchmarking lightweight NLP models on medical emergency detection
- Research on urgency language patterns in South Asian healthcare contexts

---

## ⚠️ Limitations

- **Synthetic training data** — The model was trained on AI-generated data. Real-world requests may use different phrasing, slang, abbreviations, or informal language not well-represented in training.
- **English only** — Sri Lankan blood requests often appear in Sinhala or Tamil. This model does not support those languages.
- **Sri Lanka context** — The dataset references Sri Lankan hospitals and cities. Performance on requests from other regions may vary.
- **Not for clinical use** — This model must not be used as a substitute for medical triage or clinical decision-making. It is an assistive tool for a donor coordination platform only.

---

## 🚀 Quick Start

### Using the Pipeline (Recommended)

```python
from transformers import pipeline

classifier = pipeline("text-classification", model="AshenFdo/emergency_blood_request_classifier")

result = classifier("O negative blood required urgently at Karapitiya Hospital. We are out of time.")
print(result)
# [{'label': 'emergency', 'score': 0.999...}]

result = classifier("Organizing AB+ blood for an upcoming planned operation at Kandy National Hospital.")
print(result)
# [{'label': 'not_emergency', 'score': 0.999...}]
```

### Loading Model Directly

```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("AshenFdo/emergency_blood_request_classifier")
model = AutoModelForSequenceClassification.from_pretrained("AshenFdo/emergency_blood_request_classifier")

text = "EMERGENCY: Kandy National Hospital ICU urgently needs A- blood. Patient is critical."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)

with torch.no_grad():
    logits = model(**inputs).logits

predicted_class_id = logits.argmax().item()
label = model.config.id2label[predicted_class_id]
print(f"Prediction: {label}")
# Prediction: emergency
```

### Batch Inference

```python
from transformers import pipeline

classifier = pipeline("text-classification", model="AshenFdo/emergency_blood_request_classifier")

requests = [
    "Need O- blood for a routine hospital visit at Negombo Hospital.",
    "My sister is in Kandy Hospital ICU with organ failure. We desperately need B+ blood. Pls help!",
    "Community blood donation drive at Colombo town hall this Saturday.",
    "URGENT: Karapitiya hospital needs AB- blood. Bus accident, multiple casualties.",
]

results = classifier(requests)
for req, res in zip(requests, results):
    print(f"[{res['label']}] {req[:60]}...")
```

---

## 📊 Training Details

### Dataset
- **2,500** synthetic blood request descriptions
- **Balanced** — 1,250 `emergency` and 1,250 `not_emergency`
- Sri Lanka-specific hospital names and geographic references
- See the full dataset: [AshenFdo/synthetic_blood_request_urgency_dataset](https://huggingface.co/datasets/AshenFdo/synthetic_blood_request_urgency_dataset)

### Label Definitions

| Label | Description |
|---|---|
| `emergency` | Patient is in a critical or life-threatening condition requiring blood immediately. Typically includes keywords like "urgent", "critical", "out of time", "severe hemorrhage", ICU references, or accident/trauma scenarios. |
| `not_emergency` | Routine, planned, or replacement donation requests. Includes scheduled surgeries, chronic condition support, community blood drives, and replacement donor programs. |

### Hyperparameters

| Parameter | Value |
|---|---|
| Learning Rate | `1e-4` |
| Train Batch Size | `32` |
| Eval Batch Size | `32` |
| Epochs | `10` |
| Optimizer | AdamW (fused) |
| LR Scheduler | Linear |
| Seed | `42` |

### Training Results

| Epoch | Training Loss | Validation Loss | Accuracy |
|---|---|---|---|
| 1 | 0.0954 | 0.0107 | 0.998 |
| 2 | 0.0041 | 0.0010 | 1.000 |
| 3 | 0.0109 | 0.0017 | 0.998 |
| 4 | 0.0001 | 0.0001 | 1.000 |
| 5–10 | ~0.0000 | ~0.0000 | 1.000 |

### Framework Versions
- Transformers: 5.10.1
- PyTorch: 2.11.0+cu128
- Datasets: 5.0.0
- Tokenizers: 0.22.2
- Training Environment: Google Colab (GPU)

---

## 🗂️ Example Predictions

| Description | Expected Label |
|---|---|
| *"HOSPITAL EMERGENCY: Ratnapura Hospital urgently requires AB- blood for a critical patient."* | `emergency` |
| *"O negative blood required urgently at Karapitiya Hospital. We are out of time."* | `emergency` |
| *"Need O+ blood ASAP. Friend is in Ragama hospital surgical ICU with a burst spleen."* | `emergency` |
| *"Organizing AB+ blood for an upcoming planned operation at Kandy National Hospital."* | `not_emergency` |
| *"Blood donation campaign at the Ratnapura clock tower organized by the local association."* | `not_emergency` |
| *"Advance donor support for an elective surgery at Ratnapura Hospital. A- needed."* | `not_emergency` |

---

## 🔗 Related Resources

- 📁 **Full Project on GitHub:** [AshenFdo/Blood-Request-Emergency-Classification-Model](https://github.com/AshenFdo/Blood-Request-Emergency-Classification-Model)
- 🗃️ **Training Dataset:** [AshenFdo/synthetic_blood_request_urgency_dataset](https://huggingface.co/datasets/AshenFdo/synthetic_blood_request_urgency_dataset)
- 🤗 **Base Model:** [distilbert/distilbert-base-uncased](https://huggingface.co/distilbert/distilbert-base-uncased)

---

## 📜 Citation

If you use this model in your research or project, please cite it as:

```bibtex
@model{AshenFdo_emergency_blood_request_classifier_2025,
  author    = {AshenFdo},
  title     = {Emergency Blood Request Classifier},
  year      = {2025},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/AshenFdo/emergency_blood_request_classifier}
}
```

---

*Built with the goal of saving lives — one donation at a time. 🩸*