Using Machine Learning models to effectively predict heart attacks before they happen using data easily obtainable from a standard doctor's appointment
| Precision | Recall | Accuracy | F1 | F2 |
|---|---|---|---|---|
| 82.61% | 95.00% | 91.53% | 0.88 | 0.92 |
The dataset has 294 rows, with 14 features. Several of the values are missing, and are marked with ?.
The dataset can be found in heart-attack-prediction.csv.xz in this repository. Use the command xz -d *.csv.xz to decompress the archive. It is also available from Kaggle here.
Models were trained on all samples with a train-test split of either 80/20 or 90/10. The metrics evaluated were precision, recall, accuracy, F1 score, and F2 score.
The decision was made to choose the model via Recall, since in this application a false negative puts the individual at risk due to not receiving the required treatment.
I worked on this project as a part of my summer internship as a Data Analyst at pSolv. I presented it to the founders of the company, and we are planning to pitch it to the international hospital chain Christus, who we have worked with in the past. Here is the presentation I used. If our proposal is successful, I will be working with sensitive data, so I will not be able to publish anything here.
For more information please contact me at arjvik@gmail.com.