Improvement of a tool to detect technical or health-related events at the scale of several groups of sows digital agriculture has grown considerably, especially in pig farming. It is currently possible to track animals individually and collect information that can help farmers make decisions. The objective of this study was to test and compare three methods of event detection (technical or health-related) from temporal data (water consumption and physical activity) collected by automatons and sensors. The long-term goal is to successfully develop the most reliable automated tool to quickly identify changes in drinking behaviour and/or physical activity that could indicate a technical or health-related event. Data were collected from 63 pregnant sows at the INRAE UE3P experimental facility at Saint-Gilles. Four groups of sows experienced either two thermal disturbances or two dietary disturbances. In this study, the three methods of event detection tested and compared were 1) differential smoothing, 2) detection of breaks in correlations between different pairs of variables, and 3) random forests. The best detection of thermic or dietary disturbances was obtained with random forests (i.e. for one group, sensitivity to the detection of a cold event of 53% and positive predictive value of 75%). To improve detection performance, intra-and inter-individual variability should be considered better by establishing animal-specific reference curves during complete gestation periods with no events.