These terms are commonly used in statistics, particularly in the fields of epidemiology and machine learning, to evaluate the performance of a model or a test. Here are brief explanations of each term:
- Accuracy: This is the proportion of true results (both true positive and true negative) in the total number of cases examined. To put it simply, it is the number of correct predictions made by the model over the total number of predictions. Accuracy is a useful measure when the target variable classes in the data are nearly balanced.
Formula: (True Positives + True Negatives) / (True Positives + False Positives + True Negatives + False Negatives)
- Sensitivity (also known as True Positive Rate or Recall): This is the proportion of actual positive cases which are correctly identified as such (e.g., the percentage of sick people who are correctly identified as having the condition).
Formula: True Positives / (True Positives + False Negatives)
So if a test has a high sensitivity and returns a negative result, you can be fairly sure that the condition being tested for is not present (high “negative predictive value”).
- Specificity: This is the proportion of actual negative cases which are correctly identified as such (e.g., the percentage of healthy people who are correctly identified as not having the condition).
Formula: True Negatives / (True Negatives + False Positives)
If a test has a high specificity and returns a positive result, you can be fairly sure that the condition being tested for is present (high “positive predictive value”).
It’s important to note that no test is 100% sensitive or specific, meaning that false positive and false negative results will always exist. In practice, we often have to decide which we worry about more (false positives or false negatives) and then choose a test that performs best by that criterion. Moreover, a balance between these parameters is often sought depending upon the context or the severity of the outcome of the false predictions.