Activity classification in sports is a powerful tool for athlete monitoring, enhancing performance and injury prevention. In handball, detection and classification of throws during a practice or a (practice) match has not been done. Therefore, the aim of this study is to use machine learning algorithms to detect handball throws and to classify between different throwing types and wind ups in handball based on wrist IMU data during a practice or practice match. A total of 2475 throws from 16 players were used for the detection and classification. Multiple algorithms were tested for the binary (throw versus no throw) event detection. The k-Nearest Neighbours algorithm provided the highest accuracy and F1 score and is therefore the best fit. For classification, all throws were labelled with one of the 17 throw types. Five categories were made to test on what scale the classification is possible. The categories consisted of all 17 throw types, shots versus passes, wind up type, a 7-class category and an intensity-based category. Even though multiple algorithms were tested, for all categories Support Vector Machines gave the highest F1-score and accuracy and was therefore the best fit. The categories based on intensity and wind up type scored higher than the categories with all 17 throw types and with 7 classes. Future research should focus on balancing out and enlarging the dataset, preferably with lab data.