This study investigates if a smartphones orientation in the pocket affects the result of a decision tree model trained with data from personal falls, and also how a low-pass filter affects these results. A comparison is made between the results gathered from this study, compared to previous studies and products within the field. The data was gathered using a smartphone application and was later split up to get datasets for all the different orientations of the smartphone. Before training the models, the data was processed through a low pass filter. Results showed that low pass filtered signals generally performed better and that two of the trained models, could outscore at least one other algorithm cited in this thesis in at least one category. However, existing products on the market that were investigated do not disclose their statistics and a comparison to these products could not be made. The best two orientations for the phone to be placed in the pocket was when the face of the phone was pointing out from the leg, and top of the phone was pointing up and also when the face of the phone was pointing out from the leg, and the top of the phone was pointing down.