Machine learning of anonymous call data from national suicide prevention helpline services

: understanding caller behaviour and policy implications

Abstract

Telephone crisis helplines, operated by trained volunteers/counsellors, provide support to individuals experiencing a crisis and are at risk of suicide. Callers are assisted in ways to de-escalate their crisis. Calls are logged digitally on telephone call logs, providing opportunities for data analytic and machine learning techniques to uncover insights about caller behaviour. This research presents studies on call log data provided by numerous crisis helpline services to gather insights on caller behavior which can inform crisis helpline operations. Call data from a crisis helpline is subjected to analytic methods, uncovering the presenting reasons for callers contacting the helpline, with suicide related themes being the most prevalent. Intra-call event data is applied to process mining to illustrate the caller experience. The impact of a regional routing mechanism to branches across different regions of the United Kingdom from a national crisis helpline with intraregional calls lasting longer in duration than interregional calls. Attributes of caller behaviour are subjected to k-means clustering which uncovered five caller archetypes. Caller management measures applied by a national crisis helpline were examined on their impact on the five caller archetypes, with changes to the behaviour of the most frequent callers. The impact of tenure i.e. the period between first and last call to the service, was examined on clustering of caller behaviour and found that the caller archetypes remained stable when controlling for tenure. Time series forecasting methods are compared in their ability to forecast future call arrivals to the service over different time periods, with varying results. The impact of the COVID-19 pandemic on crisis helpline caller behaviour is examined, with callers making longer calls to the service compared to when COVID-19 is a non-factor and with temporal changes in caller behaviour. Implications for policy and practice are discussed along with prospective future research.

Date of Award	Oct 2021
Original language	English
Sponsors	Department for the Economy & Ulster University
Supervisor	Edel Ennis (Supervisor), Siobhan O'Neill (Supervisor), Maurice Mulvenna (Supervisor) & Raymond Bond (Supervisor)