Personalized activity recognition, which targets specificsingle-person in multi-person environments, can be veryeffective in various settings where each person has their own objectsand movement patterns. However, most activity recognitionresearches only deal with general activity recognition, which usesthe same model for all single-person individuals. This is becauseit is difficult to build a customized model for each individual viamanual feature engineering. Thus, in this paper, we introducepersonalized activity recognition as a new research direction andpropose our own approach to build model for each individualusing recurrent neural network (RNN). Also, we suggest a graphbasedevent processing approach to seamlessly collect time-slicedand annotated data. Finally, we construct three kinds of RNNarchitectures with three different unit types including iRNN, LSTM and GRU, and perform experiments using real dataset. From the experimental results, we conclude that our approachis feasible to build the customized model in real-world forpersonalized activity recognition.