JMIR Publications

ABSTRACT

Background:

Domestic Violence (DV) is one of the most underreported public health issues. Leveraging various sources for assessing victims’ experiences is indispensable; however, traditional qualitative analysis and/or statistical analysis limited at incorporating the diverse aspects of victims’ experiences. With the unprecedented surge of data available on the internet and clinical systems, leveraging machine learning (ML) to detect obscure changes and predict the likelihood of DV from digital text data is particularly germane to advance the health science research. Researchers leveraged ML methods to examine DV from various data sources and perspectives. However, there is a paucity of research done on discussing and reviewing ML applications in the realm of DV.

Objective:

The objective of this review was to identify the current use of ML and the implementation challenges in DV.

Methods:

We extracted 3,176 articles from four databases. Twenty-two articles met the inclusion criteria. Data retrieved from the articles included the ML method (i.e., supervised or unsupervised), study group, data source, use of ML, challenges in conduction and implementation.

Results:

Twelve articles used supervised method, seven articles used unsupervised ML method, while three articles applied both. The study group dominated in DV population (n=13), partially presented in intimate partner violence (n=5), gender-based violence(n=1), child abuse (n=2), mixed DV and child abuse (n=1). Data source included social media, professional notes, national database, survey and newspaper. Random Forest(n=9), Support Vector Machine (n=8), and Naïve Bayes (n=7) are the top three algorithms for supervised ML training, while top used automatic algorithm for unsupervised ML in DV research is Latent Dirichlet Allocation (LDA) for topic modeling (n=2). The purposes of ML were identified: exploration of themes or hidden topics, prediction of future crime, and classification of the DV likelihood. Sixteen studies discussed the challenges, including data source reliability and availability, long data pre-processing time, end user acceptance, professional compliance, and professional training to handle the dataset.

Conclusions:

Leveraging ML method to examine DV has been advanced exponentially in the past decade. However, further research is needed to examine the adoption difficulties in different settings and compare the ML effectiveness with the traditional method. Clinical Trial: N/A