lemonella/surgery-match-predictor

Vision Surgery Match Predictor

Project Overview

A local machine learning model that will learn requirements for being accepted for surgery and will make decisions if the applicant is a match.

It will learn from previous filled out forms with the decision, find patterns of why it was or wasn't accepted and learn to make the decision itself. ---

1. Tech Stack

• —Language: Python
• —Machine learning framework: scikit-learn
• —Frontend form framework: streamlit
• —Data processing: pandas, numpy
• —Project version control: Gitlab

Features

• —Completely Local
• —Generate match propability in percent
• —Add a reason to the decision
• —Point out potential problems

• —UI for filling in form, submitting it and getting a percentual surgery eligibility and reasons

Timeline

June 26

• —Discuss project idea and requirements

June 29

• —Got data and explanation of data structure

June 30

• —Started to get to know the table structure and different issues

July 1

• —Created a data inspection pipeline that creates a .csv file and decides if the patient was accepted or not based on the data in the table

July 2

• —Started to create a model pipeline that will train a model based on the data

July 3

• —Started to create a prediction model that will take in a new patient and predict if they are a match or not

July 6

• —Indentifying the models mistakes and upgrading the pipeline to cover some issues

July 7

• —Making the model more accurate by ignoring some that look like mistakes / trolls

July 8

• —Creating a model pickle, that can be used to predict, or trained to be better
• —Made a training pipeline to train the model on the data

July 9

• —Developing the prediction model to be able to take a database entry and predict if the patient is a match or not

July 10

• —Finishing up the prediction model

July 13

• —Making a frontend form with the questions with a percentual match and a reason for the decision after submitting
• —Making the training pipeline cover more problems and handle the different issues more balanced and reasonable

July 14

• —Making the frontend more like the original
• —Making the perentage change in real time while filling in the form

July 15

• —Tweaking the model to handle more edge cases and make it more realistic

July 16

• —Adding simulated data to the pipleine to influence the decision making of the model

July 17