The alarm system plays a vital role to ensure safety and reliability in the process industry. Ideally, an alarm should inform the operator about critical conditions only and provide guidance to a set of corrective actions associated with each alarm. During alarm floods, the operator may be overwhelmed by several alarms in a short time span, and crucial alarms are more likely to be missed during these situations. Most of the alarms triggered during a flood episode are nuisance alarms –i.e. alarms that do not convey any new information to the operator, or alarms that do not require operator actions. Chattering alarms that repeat three or more times in a minute and redundant or duplicated alarms are common forms of nuisance alarms. Identifying such nuisance alarms is a key step to improve the performance of the alarm system. Recently, advanced techniques for alarm management have been developed to quantify alarm chatter; although effective, these techniques produce relatively static results. Machine learning algorithms offer an interesting opportunity to analyse historical alarm data and retrieve knowledge, which can be used to produce more flexible and dynamic models, as well as to predict alarms behaviour. The present study aims to develop a machine learning-based algorithm for chattering prediction during alarm floods. A modified approach based on run lengths distribution has been developed to evaluate the likelihood of future alarm chatter. The method has allowed categorizing historical alarm events as alarms that will (or will not) show chattering in the future. Finally, categorized alarms have been used to train a Deep Neural Network, whose performance has been evaluated against the ability to predict alarm chatter. Overall, the Neural Network has shown good prediction capabilities and most of the chattering alarms were correctly identified.