Over the years, Human Occupancy Measurement has had and continues to have a faire share of attention by both the research and industry communities. This long-term interest has been supported by the recent technological advances, such as the emergence of the Internet of Things (IoT), which offers a cheap alternative for gathering and processing various environmental streams of data closer to the edge, as well as machine learning techniques capable of crunching considerable amounts of raw data in real-time to produce useful and meaningful information. This paper explores and discusses the performance of a selection of machine learning algorithms applied on non-intrusive environmental sensor data (temperature and humidity) in order to infer human occupancy in closed office spaces. This work serves as a framework to help both researchers and practitioners get a clearer idea on the efficiency and performance of each algorithm in terms of accuracy, precision, as well as other metrics. It also provides a walkthrough of time series data handling and preparation in the context of office occupancy detection. The results are also compared to a solution relying on classic data analysis methods requiring expert knowledge of the problem.