Predicting the Severity of Apnea through Machine LearningData Science and Math
- Kevin Nieves Pichardo
Prediction modeling has limitless potential in insurance, retail, and healthcare. Predictive models allow corporations and health officials to make decisions based on statistical data. Obstructive sleep apnea (OSA) is a severe disease that affects 5% to 14% of adults. This disease has caused adverse effects to its constituents, and most people suffering from severe OSA die. This project presents a prediction model using random forest (RF) to accurately predict patients with none and severe OSA to make it easier to diagnose. Previous works have addressed models for prediction of OSA severity, but no significant study has been conducted using random forest model. This process was done by collecting demographic and polysomnography (PSG) data from 781 participants. Data analysis was conducted by only considering none and severe OSA and removing those with not applicable variables leading to only 339 participants being considered for model. The data was initiated in the RF model through the program RStudio. The results equated to 74.34% accuracy based on the confusion matrix for testing data. Making RF prediction model a valid model to predict OSA, this can shorten the diagnosis timeframe from a week to just one day, as the model can make its predictions through a questionnaire and overnight PSG test. This quick analysis allows sleep specialists to diagnose and treat the patient quicker.