Purpose of resampling

A hospital is developing a machine learning model to predict whether patients will develop a rare disease based on their medical records.

However, in the hospital's historical data, only 5% of patients were diagnosed with the disease, while 95% were healthy. When testing an initial model, it achieved 95% accuracy, but it rarely predicted the disease, meaning it was mostly just predicting "healthy" for everyone.

You're consulting for the hospital and advised to apply synthetic resampling. What is your main argument to apply resampling in this case?