Fig. 3: Prediction-report lead time in self-report patients.
From: Leveraging multimodal machine learning for accurate risk identification of intimate partner violence
The prediction-report lead time is calculated as the difference between the patient’s first self-report date and the earliest date when the model detected IPV; a positive result indicates early detection. We compare this value to the earliest possible time gap: the difference between the first time the patient encounters the healthcare system and when they report IPV. Both the tabular model (a) and the fusion model (b) can detect IPV risk on average more than three years in advance. While the tabular model has slightly higher mean and median prediction lead times, the fusion model can detect more IPV cases in advance.
