The Limitation of Accuracy Score
Accuracy is a popular metric for machine learning evaluation, but there is some limitation to the metrics if you use them as the only metric. The limitation is:
Accuracy is biased when there is an imbalance of data. The accuracy calculation would favor the majority class as the model that keeps predicting the majority class would have great accuracy.
Accuracy took the false prediction as equal to each other, whether a false positive or false negative. In reality, some misclassification cost more than others. For example, a false negative in cancer prediction is more dangerous than a false positive.
Accuracy is not a good representation for models with multiclass labels. Similar to the imbalance cases, a model that consistently predict a majority label on the multiclass would always have a good score.
That is why there are still various metrics to evaluate our models, such as Precision, Recall, or F1 score.
That is all for today; thank you for subscribing, and let’s meet on the comment.