Negative Predictive Value (NPV) is a statistical measure used in the context of diagnostic testing and modelado predictivo. It quantifies the proportion of true negative results among all negative test outcomes. In simpler terms, NPV helps determine how reliable a test is when it indicates that a subject does not have a certain condition or trait.
Matemáticamente, el VPN se define como:
VPN = Verdaderos Negativos / (Verdaderos Negativos + Falsos Negativos)
Donde:
- Verdaderos Negativos (TN) are the instances where the test correctly identifies the absence of the condition.
- Falsos Negativos (FN) are the instances where the test incorrectly indicates the absence of the condition when it is actually present.
El VPN es particularmente útil en entornos clínicos, donde puede informar healthcare professionals about the likelihood that a patient actually does not have a condition based on their test results. For instance, a high NPV indicates that a negative result can be trusted to mean the patient is likely healthy, while a low NPV suggests that negative results could be misleading, warranting further investigation.
In analítica predictiva and aprendizaje automático, NPV becomes crucial when evaluating classification models, especially in scenarios where the prevalence of the condition is low. In these cases, models may produce a high number of negative predictions, making NPV an essential metric for entender el rendimiento del modelo.