Sleep disorders occur in around 20% of the population and are associated with a multitude of serious health implications. Whilst polysomnography is considered the ‘gold standard’ in assessing sleep patterns, it is intrusive and costly hence actigraphy is increasingly being considered as a viable alternative. Current actigraphic sleep pattern detection algorithms typically use filtered uni-axial actigraphy data with a low sampling rate, measured on a patient’s wrist or hip. However, the utility of wrist algorithms are limited due to poor specificity (wakefulness detection) and the ability for hip actigraphy to monitor sleep patterns is relatively unsubstantiated. Using simultaneously measured Actigraph GT3X+ and polysomnography data from 100 healthy young adults (22 years old) in the Western Australian Pregnancy Cohort (Raine) Study we created and validated two sleep/wake identification algorithms. These algorithms utilized raw tri-axial acceleration data from wrist and hip actigraphy. Random forest methodology was used as a form of dimension reduction and subsequent generalized linear mixed models were used to form predictive models. Our final wrist actigraphy algorithm yielded an average accuracy of 74% (SD 17%), sensitivity (sleep detection) of 75% (SD 18%) and specificity of 69% (SD 22%). The final hip actigraphy algorithm yielded an average accuracy of 70% (SD 13%), sensitivity of 71% (SD 15%) and specificity of 65% (SD 22%). These results were superior in specificity and comparable in accuracy and sensitivity to other widely used algorithms, indicating that raw data is suitable for sleep/wake classification. In addition, we found our model derived from hip actigraphy was comparable to the model derived from wrist actigraphy, supporting the notion that hip actigraphy could be an effective tool for sleep/wake identification.