High confinement tokamak fusion devices are characterized by magnetohydrodynamic instabilities that occur near the edge of the plasma, called edge-localized modes (ELMs). While not necessarily catastrophic, the consequent losses in temperature and energy, as well as the wear and tear to the wall and plasma-facing components pose a significant risk. Previously, a robust peak detection algorithm was developed to find ELMs, extract their properties (timing, losses, etc.), and predict their behavior on a range of tokamaks, such as JET, ASDEX Upgrade and DIII-D, with minimal manual tuning. Recently, modes of operation with smaller, but more irregular ELMs have been gathering interest. However, detecting these erratic ELMs consistently with existing tools has been challenging. Recently, machine learning methods such as support vector machines and 1D convolutional neural network classifiers have been used to recognize anomalous events. Inspired by these tools, new ELM detection methods are proposed and their performance is compared with the existing algorithm.