Prescription Insight

Team/Project Name: Prescription Insight

Team Members: 

• Mark Lloyd T. Cuizon

The Problem    

     The challenge in healthcare is accurately assessing the risk of drug addiction in patients undergoing drug treatment. Traditional methods may not fully capture the complexity of individual cases, which can lead to less effective treatment plans (National Institute on Drug Abuse, 2020).

The Solution   

    This project introduces an AI model enhanced with GizaTech's Orion Framework, designed to offer a secure and verifiable prediction of drug abuse risk. The model uses demographic data, personality scores, and historical drug use data from the "Drug_Consumption.csv" dataset. While various models like Logistic Regression, Random Forest Classifier, SVM, and KNN were explored, the final model was developed using Logistic Regression implemented in PyTorch (focuses more on Ketamine prediction). This approach facilitates the transpilation of the model into Orion's framework, ensuring enhanced security and verifiability of the machine learning algorithms and the data used.

Value Proposition     

     By integrating GizaTech's Orion Framework, this AI model brings a new level of security and trustworthiness to risk assessments in drug addiction. It provides healthcare professionals with a tool that not only improves the accuracy of risk evaluations but also assures the protection of sensitive patient data. This advancement is crucial for personalized healthcare and promises better patient outcomes (World Health Organization, 2019).

The Technology Used    

     The project leverages state-of-the-art machine learning techniques and integrates them with GizaTech's Orion Framework and Technology. The use of Logistic Regression in PyTorch for the development of the model aligns with Orion's capabilities for creating verifiable and secure machine learning models, making it an ideal choice for healthcare applications where data security and model reliability are paramount (Hastie, T., Tibshirani, R., & Friedman, J., 2009).

Future Recommendations    

     The development of a user-friendly application for healthcare professionals, either as a Windows desktop app or a web-based solution, is recommended. Incorporating frameworks like Tkinter for desktop applications or Flutter for web-based solutions can allow medical professionals to easily input predictor variables and receive risk assessments. This application would be built on the Orion Framework, ensuring the security and verifiability of the assessments provided.

Giza Actions:

-SK-learn Logistic Model Training and ONNX conversion

P.S. Since sklearn's Logistic regression (ONNX converted model) is not yet fully supported with GizaTech's transpiler. An alternative was done by creating a Logistic Regression using PyTorch from scratch.

-PyTorch Logistic Model Training and ONNX Conversion

Transpilation Process: 

Verification of Generated ZK Proof:

Github Link: https://github.com/Markishime/Prescription-Insight