Klassificering af Kardiovaskulær sygdom: Klassificering af Kardiovaskulær sygdom

Kristensen, Daniel ; Reinholt, Tobias ; Holm, Mikkel Stouby ; Mahmoud, Salar ; Khorshid, Lanja Mozafar ; Vibæk, Mathilde

1. semester

Design og anvendelse af kunstig intelligens, Diplomingeniør

2024

2024-12-18

34 pages

This project aims to implement and test different machine learning methods for the development of a model to classify whether patients have or don’t have cardiovascular disease (CVD). The intent is to combine the model with a software product for general practitioners to use. The purpose of this product is to reduce mortality and morbidity by recommending certain patients for further assessment. It was found that certain risk factors are used to evaluate the risk of CVD. This was important in deciding which features from the dataset to include in the model. Initially three classifiers where chosen through a Quick and Dirty process where a total of six classifiers were tested, with special attention to recall and F1. SVM was implemented and tested in this project along with Random Forest and Logistic Regression. They were initially tested separately to find their optimal hyperparameters before being combined in an ensemble model. Soft voting was used and the hyperparameters were once again tuned to maximize the collective performance for the ensemble. The training showed, that the ensemble model performed the best with an accuracy of 81%, precision of 84%, recall of 84% and F1 of 83%. The optimal decision threshold was then estimated for the ensemble model at 0.376 resulting in an accuracy of 82%, precision of 77%, recall of 96% and F1 of 85%. Due to lack of time, this project did not succeed in creating a user interface for receiving user input in the form of medical data. The project fulfilled the requirement of a minimum recall of 95%, but did not achieve the required minimum accuracy of 90%. Future work on the project would focus on improving accuracy and the creation of a software product to receive user input.

Download
Vis denne rapport i AAU Studenterprojekter