Protein phosphorylation, regulated through protein kinases, is a key post-translational modification that coordinates many protein activities and functions, and mediates most cellular signalling. Some mutations can result in constitutive activation of kinases, leading to a loss of control over cellular phosphorylation, and the development of many diseases, including metabolic disorders and notably many cancers. However, the identification of activating mutations has been limited to direct experimental evidence or sequence-based comparisons with previously characterised mutations. We present a novel predictive tool, named Kinact, which uses models trained with supervised machine learning algorithms to identify kinase activating mutations. Experimental data on 384 point mutations in 42 different protein kinases was manually curated, and each kinase classified into its respective kinase group according to the Standard Kinase Classification Scheme. The structural effects of these mutations on the wild-type residue environment, interatomic interactions and protein stability and flexibility were calculated, which along with sequence information, were used as features for training supervised learning algorithms. By combining the structural based predictions with conservation within the kinase family group, Kinact was able to achieve a precision of 90% and AUC of 0.96 under 10-fold cross-validation, and a precision of 81% and AUC of 0.89 across blind tests. We show that Kinact significantly outperformed the gold-standard methods used by clinical geneticists, SIFT and PolyPhen-2 (p-value < 0.01). Kinact conveniently combines high-performance open source web visualization tools to enable the rapid characterisation of how variants are likely to affect kinase activity. The method is freely available at <http://biosig.unimelb.edu.au/kinact/>.