The aim of this study was to combine the use of IMUs in football and the principles commonly used in Human Activity Recognition, to be able to predict some of the most common football actions, and herein also which placements of the sensors are the most important. This process was split into two parts, starting with predicting the football actions, i.e. pass, dribbling, first touch, and positioning, using a bidirectional Long Short Term Memory neural network (LSTM). The second part consisted of predicting head scans with a Deep Learning Artificial Neural Network (DNN) separately, as the head scan happened simultaneously with the other actions. 14 male and 3 female football players participated in the study. Prior to any predictions, the data was split 50/50 into labeled and unlabeled data, and the labeled data was further split 80/20 into training data and testing data. All data were normalized and balanced by using Adaptive Synthetic Sampling Approach (ADASYN). 5250 statistical time domain features were calculated over a sliding window of 200 ms, with 50 \% overlap, but later reduced with a Principal Component Analysis retaining >95 \% of the variance. A semi-supervised uncertainty-aware pseudo-labeling technique was used to decrease the time needed for labeling. The LSTM showed decent results for predicting football actions, with a cross-validation score of 0.74 and an F1-score of 0.74. The DNN prediction of head scans showed overall slightly better results, mainly due to the lower number of classes, with a cross-validation score of 0.79 and an F1-score of 0.78. The sensor placement that supplied the most relevant information to the LSTM was the one placed on the right calf with an F1-score of 0.65. For the DNN the most important sensor placement was the one placed on the head, which showed an F1-score of 0.69.