The growing interest in non-invasive sleep monitoring has supported the development of methods for processing actigraphic data from wearable sensors using machine learning. The aim of this thesis is to detect sleep in the form of binary sleep/wake classification. Attention is given to the selection of a suitable dataset, the design of data preprocessing procedures, and the implementation of classification models. Three classical machine learning models and five neural models, including two custom architectures, DG-SleepNet and OAM-TCN, were trained and evaluated using selected quality metrics. The results show that OAM-TCN achieved the highest overall accuracy (accuracy = 84.9%), the random forest model best maintained the balance between correct detection of sleep and wake states (BACC = 80.4%, MCC = 61.4%), and DG-SleepNet showed the best ability to distinguish between these classes across decision thresholds (ROC AUC = 87.2%). The proposed methods achieved results comparable to current state-of-the-art approaches and confirm the usability of a single accelerometer sensor for automated sleep detection.