AbstractMachine learning techniques are now well established in experimental particle physics, allowing detector data to be analyzed in new and unique ways. The identification of signals in particle observatories is an essential data processing task that can potentially be improved using such methods. This paper aims at exploring the benefits that a dedicated machine learning approach might provide to the classification of signals in dual-phase noble gas time projection chambers. A full methodology is presented, from exploratory data analysis using Gaussian mixture models and feature importance ranking to the construction of dedicated predictive models based on standard implementations of neural networks and random forests, validated using unlabeled simulated data from the LZ experiment as a proxy to real data. The global classification accuracy of the predictive models developed in this work is estimated to be >99.0%, which is an improvement over conventional algorithms tested with similar data. The results from the clustering analysis were also used to identify anomalies in the data caused by miscalculated signal properties, showing that this methodology can also be used for data monitoring.