Jainam-11/symptom-based-disease-prediction-v1

Symptom-Based Disease Prediction Dataset (Confidence-Aware) – v1

📘 Overview

This dataset is an early-stage (Version 1) medical dataset created for symptom-based disease prediction using Large Language Models (LLMs).

Each record presents patient symptoms in an instruction-style prompt and returns multiple possible diseases grouped by confidence levels.
The primary goal of this version is to establish structure, consistency, and reasoning format, not final model performance.

⚠️ Important Note:
Version 1 is NOT fine-tuning ready.
It is a base dataset used for experimentation, validation, and iteration.
We are actively working on Version 2, which will be cleaner, more balanced, and fully suitable for LLM fine-tuning.

---

🎯 Intended Use (v1)

This version is suitable for:

• Dataset structure exploration
• Prompt–output format testing
• Research and experimentation
• Prototyping medical reasoning pipelines
• Feedback-driven dataset improvement

Not recommended yet for production-grade fine-tuning.

---

🧠 Key Characteristics

• Symptom-to-disease reasoning focus
• Multi-disease prediction (not limited to one label)
• Confidence-aware outputs:
  • high_confidence
  • medium_confidence
  • low_confidence
• Supports incomplete, overlapping, and ambiguous symptom inputs
• Instruction-style prompts for causal LLMs
• Stored in JSONL format for easy parsing and scalability

---

📂 Dataset Format

The dataset is stored in JSON Lines (.jsonl) format, where each line represents one sample.

Schema

{
  "input": "string",
  "output": {
    "high_confidence": ["string"],
    "medium_confidence": ["string"],
    "low_confidence": ["string"]
  }
}