---
title: Actuarial Skills Classifier
emoji: 🎯
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: true
license: apache-2.0
tags:
  - actuarial
  - multilabel-classification
  - sentence-classification
  - modernbert
  - skills-extraction
  - career-planning
  - job-analysis
datasets:
  - actuarial-jobs
language:
  - en
metrics:
  - f1
  - precision
  - recall
---

# 🎯 Actuarial Skills Classifier

**Discover the actuarial and programming skills you need—instantly.**

This Hugging Face Space uses **multilabel sentence classification** trained on **7,000+ real actuarial job postings** to instantly map your text to employer-demanded skills. Simply describe a task, role, or career goal, and watch relevant actuarial competencies and coding techniques appear.

## 🚀 Quick Start

Try these examples:
- *"pricing motor portfolio with GLM and Python"*
- *"life insurance reserving and IFRS 17 implementation"*
- *"building predictive models for mortality forecasting"*

## ✨ Features

### 🎯 **Skills Analysis**
- **Instant Mapping**: Type any description and get relevant skills in seconds
- **Confidence Scores**: Each skill shows match percentage (0-100%)
- **Smart Categories**: Skills organized by type (Programming, Actuarial, Tools, etc.)
- **Learning Roadmap**: Personalized priority-based development plan

### 📦 **Batch Processing**
- Upload CSV or TXT files with multiple descriptions
- Analyze up to 100 job postings simultaneously
- Identify common skills across multiple roles
- Export results for further analysis

### 🔍 **Skills Database**
- Browse all 7,000+ learned skills
- Search and filter by keywords
- Understand market demand for specific competencies

### 📚 **Comprehensive Guide**
- Detailed usage instructions
- Interpretation guidelines
- Career planning tips
- Example use cases

## 🎓 Use Cases

### **Students & Early Career Professionals**
Discover what skills you need to learn for your target role:
```
Input: "I want to become a pricing actuary in motor insurance"
Output: GLM modeling, Python, R, Motor insurance, Pricing techniques, 
        Statistical analysis, Excel, SQL, Risk assessment...
```

### **Career Transition**
Identify transferable skills and gaps when changing specializations:
```
Input: "Moving from life insurance reserving to P&C pricing"
Output: [Shows overlap + new skills needed for transition]
```

### **Job Analysis**
Paste any actuarial job description to extract all required skills:
```
Input: [Full job posting text]
Output: Complete skill breakdown with confidence scores
```

### **Learning Path Planning**
Get a structured roadmap for skill development:
```
Input: "Building machine learning models for insurance pricing"
Output: High Priority → Medium Priority → Nice to Have skills
```

### **Hiring & Recruitment**
Analyze what skills are commonly required together in the market:
```
Input: [Multiple job descriptions via batch upload]
Output: Most frequent skill combinations and trends
```

## 📊 Model Details

### Architecture
- **Base Model**: [answerdotai/ModernBERT-base](https://huggingface.co/answerdotai/ModernBERT-base)
- **Task**: Multi-label sequence classification
- **Fine-tuning**: Trained on 7,000+ actuarial job postings
- **Labels**: 100+ unique skills spanning actuarial domains, programming, tools, and soft skills

### Training Data
Our model was trained on a diverse dataset of real actuarial job postings covering:
- **Industries**: Life, P&C, Health, Pensions, Reinsurance
- **Regions**: Global actuarial markets
- **Experience Levels**: Entry-level to executive positions
- **Time Period**: Current market requirements (2023-2025)

### Performance Metrics
| Metric | Score |
|--------|-------|
| F1 Macro | 0.XXX |
| F1 Micro | 0.XXX |
| Precision | 0.XXX |
| Recall | 0.XXX |

*(Replace with actual metrics from your training)*

### Skills Coverage

The model recognizes skills across multiple categories:

**🔤 Programming Languages**
- Python, R, SQL, VBA, SAS, C++, Java, JavaScript, Scala

**📊 Actuarial Domains**
- Pricing, Reserving, Capital Modeling, Valuation
- Life, P&C, Health, Pensions, Reinsurance
- Risk Management, Product Development

**🤖 Technical Skills**
- Machine Learning, Deep Learning, GLM, GAM
- Statistical Modeling, Data Science, Analytics
- NLP, Computer Vision, Time Series

**🛠️ Tools & Software**
- Excel, Prophet, ResQ, MoSes, AXIS
- PowerBI, Tableau, Qlik
- Git, Docker, Apache Spark
- TensorFlow, PyTorch, scikit-learn

**💼 Business Skills**
- Communication, Presentation, Leadership
- Stakeholder Management, Strategy
- Project Management, Team Collaboration

**📜 Certifications**
- ASA, FSA, ACAS, FCAS
- CERA, CFA, FRM
- Regional qualifications (FIA, DAV, etc.)

## 🎯 How It Works

### 1. Input Processing
```
Your text → Tokenization (ModernBERT tokenizer) → Max 512/1024/2048 tokens
```

### 2. Model Inference
```
Tokenized input → ModernBERT encoder → Multi-label classification head
→ Sigmoid activation → Probability scores for each skill
```

### 3. Threshold Application
```
Raw probabilities → Filter by confidence threshold (default 0.5)
→ Rank by score → Return top K skills
```

### 4. Post-Processing
```
Predicted skills → Categorization → Learning roadmap generation
→ Formatted output for display
```

## 💡 Tips for Best Results

### ✅ DO:
- **Be specific** about domains: "motor insurance pricing" vs. "insurance"
- **Mention tools/languages**: "Python GLM model" vs. "statistical model"
- **Include context**: "IFRS 17 implementation" vs. "accounting"
- **Use industry terms**: "reserving", "capital modeling", "mortality tables"
- **Write 2-3 sentences** for richer context

### ❌ AVOID:
- Very short phrases (under 10 words)
- Generic descriptions: "data analysis job"
- Non-actuarial contexts
- Overly broad terms without specifics

### 🎚️ Adjusting Parameters

**Confidence Threshold:**
- **0.3-0.4**: Comprehensive (more skills, some less relevant)
- **0.5** (default): Balanced (good mix of precision and recall)
- **0.6-0.7**: Precise (fewer skills, highly relevant only)
- **0.8+**: Very strict (only very strong matches)

**Top K Skills:**
- **5-10**: Focus on core skills only
- **10-20** (recommended): Comprehensive overview
- **0 (unlimited)**: See everything above threshold

## 🔧 Technical Implementation

### Requirements
```txt
transformers>=4.44.0
torch>=2.0.0
gradio>=4.0.0
pandas
numpy
```

### Local Usage

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

model_name = "your-username/modernbert-actuarial-skills-classifier"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "pricing motor insurance with GLM and Python"
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)

with torch.no_grad():
    outputs = model(**inputs)
    probs = torch.sigmoid(outputs.logits)
    
# Get predictions above threshold
threshold = 0.5
predicted_indices = torch.where(probs[0] > threshold)[0]
predicted_skills = [model.config.id2label[idx.item()] for idx in predicted_indices]

print("Predicted Skills:", predicted_skills)
```

### API Usage

```python
from huggingface_hub import InferenceClient

client = InferenceClient()
result = client.text_classification(
    "pricing motor portfolio with GLM and Python",
    model="your-username/modernbert-actuarial-skills-classifier"
)

print(result)
```

### Batch Processing

```python
texts = [
    "life insurance reserving with IFRS 17",
    "P&C pricing using machine learning",
    "pension scheme valuation"
]

inputs = tokenizer(texts, return_tensors="pt", padding=True, truncation=True)
with torch.no_grad():
    outputs = model(**inputs)
    probs = torch.sigmoid(outputs.logits)

# Process each prediction
for i, text_probs in enumerate(probs):
    print(f"\nText {i+1}: {texts[i]}")
    top_k = torch.topk(text_probs, k=5)
    for score, idx in zip(top_k.values, top_k.indices):
        print(f"  {model.config.id2label[idx.item()]}: {score:.2%}")
```

## 📈 Model Training Details

### Data Preparation
1. **Collection**: 7,000+ actuarial job postings from multiple sources
2. **Cleaning**: Anonymization, deduplication, quality filtering
3. **Labeling**: Expert-annotated skill tags (100+ unique labels)
4. **Splitting**: 60% train, 20% validation, 20% test

### Training Configuration
```python
- Base Model: answerdotai/ModernBERT-base
- Max Sequence Length: [512/1024/2048] tokens
- Batch Size: [8/16/32] (depends on GPU)
- Learning Rate: 2e-5
- Epochs: 3
- Optimizer: AdamW
- Loss Function: BCEWithLogitsLoss (multi-label)
- Hardware: [GPU type used]
```

### Fine-tuning Process
1. **Tokenization**: ModernBERT tokenizer with dynamic padding
2. **Classification Head**: Multi-label binary classifier
3. **Training**: Cross-entropy loss with label smoothing
4. **Evaluation**: F1-macro, F1-micro, Hamming loss
5. **Optimization**: Gradient accumulation, mixed precision (FP16)

## 🎨 Customization

### Adding New Skills
To retrain with additional skills:
1. Prepare labeled data in same format
2. Update label mappings
3. Fine-tune with new data
4. Validate on held-out test set

### Adjusting for Other Domains
This architecture can be adapted for other professional domains:
- Software engineering skills
- Medical specializations
- Finance & banking competencies
- Legal practice areas

## 🤝 Contributing

Ways to improve this tool:
- Report issues or bugs
- Suggest new features
- Share feedback on accuracy
- Contribute training data
- Propose UI/UX improvements

## 📄 License

This project is licensed under the Apache License 2.0 - see the [LICENSE](LICENSE) file for details.

## 🙏 Acknowledgments

- **ModernBERT Team**: For the excellent base model
- **Hugging Face**: For hosting infrastructure and tools
- **Actuarial Community**: For feedback and validation
- **Training Data**: Sourced from public job boards and postings

## 📞 Contact & Support

- **Model Repository**: [Link to your HuggingFace model]
- **Issues**: Use GitHub Issues or HF Discussions
- **Email**: [Your contact email]
- **LinkedIn**: [Your LinkedIn profile]

## 🔗 Related Resources

### For Actuarial Students
- [Society of Actuaries](https://www.soa.org/)
- [Casualty Actuarial Society](https://www.casact.org/)
- [Institute and Faculty of Actuaries](https://www.actuaries.org.uk/)

### For Learning Skills
- **Python**: [Real Python](https://realpython.com/), [DataCamp](https://www.datacamp.com/)
- **R**: [R for Data Science](https://r4ds.had.co.nz/)
- **Actuarial Science**: [Coaching Actuaries](https://www.coachingactuaries.com/)
- **Machine Learning**: [Fast.ai](https://www.fast.ai/), [Coursera](https://www.coursera.org/)

### Technical Documentation
- [Transformers Documentation](https://huggingface.co/docs/transformers)
- [ModernBERT Paper](https://arxiv.org/abs/[paper-id])
- [Multi-label Classification Guide](https://huggingface.co/docs/transformers/tasks/sequence_classification)

## 📊 Statistics

- **Total Skills**: 100+ unique labels
- **Training Samples**: 7,000+ job postings
- **Model Size**: ~140M parameters
- **Inference Speed**: <1 second per prediction
- **Languages Supported**: English (with potential for multilingual expansion)

## 🚀 Future Enhancements

- [ ] Skill clustering and visualization
- [ ] Trend analysis over time
- [ ] Salary correlation insights
- [ ] Course recommendations per skill
- [ ] Multi-language support
- [ ] Mobile app version
- [ ] API endpoint for integration
- [ ] Chrome extension for job sites

## 📝 Citation

If you use this model or tool in your research or application, please cite:

```bibtex
@misc{actuarial-skills-classifier,
  author = {Your Name},
  title = {Actuarial Skills Classifier: Multi-label Classification for Career Planning},
  year = {2025},
  publisher = {HuggingFace},
  journal = {HuggingFace Spaces},
  howpublished = {\url{https://huggingface.co/spaces/your-username/actuarial-skills-classifier}}
}
```

---

<div align="center">

**Built with ❤️ for the actuarial community**

[🚀 Try the Demo](https://huggingface.co/spaces/your-username/actuarial-skills-classifier) | 
[📖 Documentation](https://huggingface.co/your-username/modernbert-actuarial-skills-classifier) | 
[💬 Discussions](https://huggingface.co/spaces/your-username/actuarial-skills-classifier/discussions)

</div>