<ul><li>Accuracy is a measure of predictive performance.</li><li>While the accuracy is straightforward, it can give a wrong idea of a model’s performance in many cases.</li><li>To avoid this pitfall, data scientists prefer another metric to measure their model performance, called the f1-score.</li><li>The f1-score is good for assessing models when there are few positives.</li><li>However, the f1-score is an asymmetrical measure and can be very misleading when the proportion of successes is very high.</li><li>A better solution is actually to use another metric called the p-4 score, which is considered the symmetrical expansion of the f1-score.</li><li>The p4-score uses the harmonic mean of the precision, recall, specificity, and negative predictive value.</li><li>The p-4 score is a very useful metric that assesses the performance of models better than the accuracy and the f1-score.</li><li>The p4-score consistently gives more sensible results than other metrics.</li><li>It fixes the main drawback of the f1-score and makes this metric far less misleading.</li></ul>

Better than the f1-score, discover the p-4 score

Discover more