Every Machine Learning model needs to be evaluated against some metrics to check how well it has learnt the data and performed on test data. These are called the Performance Metrics and are different for regression and classification models.
The most straightforward metric for classification is the Accuracy score, which just tells you how many labels are classified accurately or correctly.
However, you cannot just use accuracy for all types of classification problem. Defining a metric purely depends on the type of data/label and domain.
If your classes are skewed, accuracy score will be totally misleading. You need to know which value you need to optimize to what extent- True Positives or True Negatives or both. In this case Precision or Recall might be the right choice.
There are a lot more metrics pre-defined which I’ll be talking about in my coming posts. Having said this, it’s not just these default metrics that can be used. Custom metrics can also be defined and built which perfectly suits your data and the domain of problem you’re optimizing.
In fact, defining and building correct metrics which quantify the success in the best possible way is an utmost crucial step in the beginning of any Data Science project.
#datascience #machinelearning