The real-time prediction for the requirement of an air-ambulance (helicopter) response is very helpful in improving patient outcomes for emergency medical services (EMS). Generally, experienced and qualified Intensive Care Paramedics evaluate each incident call manually. However, during busy times with high volumes of calls, evaluating all potential incidents in a timely manner may be difficult. In this paper, we present the use of Machine Learning approach for Air ambulance prediction on an EMS dataset provided by St John New Zealand. This will give an indicator for each incident on the probability the incident requires a helicopter response, which can then be incorporated into real time information support provided to Air Desk Paramedics. In this case study, Random Forest was selected which showed promising results against some other methods, and has a potential to gain a better performance by doing feature selection and parameter optimization. This method was tested to have almost 94% classification accuracy.