A Machine Learning Approach for Predicting Non-Suicidal Self-Injury in Young Adults

Abstract

Artificial intelligence techniques were explored to assess the ability to anticipate self-harming behaviour in the mental health context using a database collected by an app previously designed to record the emotional states and activities of a group of subjects exhibiting self-harm. Specifically, the Leave-One-Subject-Out technique was used to train classification trees with a maximum of five splits. The results show an accuracy of 84.78%, a sensitivity of 64.64% and a specificity of 85.53%. In addition, positive and negative predictive values were also obtained, with results of 14.48% and 98.47%, respectively. These results are in line with those reported in previous work using a multilevel mixed-effect regression analysis. The combination of apps and AI techniques is a powerful way to improve the tools to accompany and support the care and treatment of patients with this type of behaviour. These studies also guide the improvement of apps on the user side, simplifying and collecting more meaningful data, and on the therapist side, progressing in pathology treatments. Traditional therapy involves observing and reconstructing what had happened before episodes once they have occurred. This new generation of tools will make it possible to monitor the pathology more closely and to act preventively.

Keywords: EMA; NSSI; app; machine learning.

Figure 1

On the left, a screenshot with the list of emotions to report to the system. On the right, a detail on how to report the grade of the emotion by means of a sliding button (the text in the App is in Catalan. See the translated names of the emotions in Table 1).

Figure 2

Representation of the number of interactions that each subject in the study had with his/her application. The proportion of these messages that were not directly linked to any self-harm is shown in blue, and that which was directly linked to self-harm in red. Note that of the 64 subjects present, 24 reported one or more self-injuries.

Figure 3

Representative example of the confusion matrix with bar shapes, used for the joint representation of all users’ data. The sections in the vector are proportional to the percentage of the cell corresponding to the confusion matrix. The height of the bar will be proportional to the number of interactions made by the subject. Note the colour coding to represent the TP (True Positives), the FP (False Positives), the TN (True Negatives) and the FN (False Negatives).

Figure 4

Representation of the results of the Leave-One-Subject-Out Cross-Validation test in terms of the vectorised confusion matrices scaled by the proportion of entries each subject made in the application. The vertical axis represents the number of entries for each user. We use the colour convention from Figure 3 to represent the TP (True Positives), the FP (False Positives), the TN (True Negatives) and the FN (False Negatives).

Figure 5

The positive part of the graph corresponds to the representation of the database in terms of the number of entries labelled as positive and negative for each user. The negative part of the graph corresponds to the TPs and TNs obtained in the Leave-One-Subject-Out Cross-Validation test.