shkumar0511/insurance-lob-classifier

🧠 Insurance LOB Classification Model

Model Details

Model Description

This model is a fine-tuned transformer model for multi-class classification of Insurance Line of Business (LOB) from textual data.

The model is designed for insurance, legal, and underwriting workflows where documents need to be automatically categorized.

---

• Developed by: Shubham kumar
• Model type: Transformer-based sequence classification
• Language(s): English
• License: Apache 2.0
• Finetuned from: distilbert-base-uncased

---

Intended Use

Direct Use

• Classify insurance-related text into LOB categories
• Automate document routing in underwriting pipelines
• Assist in triaging insurance submissions

Downstream Use

• Integrated into APIs for real-time classification
• Used in batch pipelines for document processing
• Combined with OCR systems for PDF ingestion

---

Out-of-Scope Use

• Not suitable for:
  • Non-insurance domains
  • Non-English text
  • Highly ambiguous or very short text
  • Legal decision-making without human review

---

Bias, Risks, and Limitations

• Model performance depends heavily on training data quality
• May be biased toward dominant classes if dataset is imbalanced
• Predictions with low confidence should be reviewed manually

Recommendations

• Use confidence threshold (e.g., 0.7) for auto vs manual routing
• Periodically retrain with new data
• Monitor class-wise performance

---

How to Use

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "shkumar0511/insurance-lob-classifier"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

def predict(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)

    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.softmax(outputs.logits, dim=1).numpy()[0]

    return probs.argmax(), probs.max()