This thesis explores what effect preprocessing has on machine learning models when training the models on ambulance data from SOS Alarm combined with weather data from SMHI. The purpose of this thesis is to more accurately estimate the ambulance arrival time by preprocessing the ambulance data and the effect of adding weather data as well as temporal data. 

In the preprocessing of ambulance data, different amount of data in the ambulance datasets is classified as outliers. The dataset is then combined with weather and temporal (date and time) features which results in multiple datasets. These datasets are then used to train three machine learning models: random forest, linear regression and artificial neural network, to measure the impact of these features on model performance. 

This thesis find that weather data has no to slight negative impact on the performance and that temporal features has no to slight positive impact on model performance. Furthermore shows that removing at least 2% of the outliers from the ambulance dataset yields significant improvement to model performance. 

The model that performed the best for the entire dataset, and the subset that contains only ambulance transports between two hospitals was the artificial neural network.For the subset that contained only the ambulance transports between Lund hospital and Malmö hospital, the best performing model was the random forest model.