How to Evaluate Your ML Models Effectively? | Evaluation Metrics in Machine Learning!

Key Takeaways

how do you evaluate your ml models evaluation is a critical step in the model development process it ensures that our model is good enough to perform well on unseen Data before diving into evaluation metrics remember that the data should be divided into train test and often validation sets more on this here let's call one class positive and the other one negative we can arrange the predictions in four possible ways either we predict the positive class correctly or we predict the negative class incorrectly and either we predict correctly the negative class or incorrectly the positive one this is called the confusion Matrix a popular metric used in classification tasks from the confusion Matrix we can extract other metrics like accuracy accuracy measures how often a model predicts correctly out of all the predictions it made if we translate that to a formula we will need to divide the total number of correct predictions by the total number of predictions let's say if a person has a certain flu we predict positive if not then negative the flu is very rare and of 100,000 people only 100 have it our model learned to classify all observations to the negative class meaning none of them has the flu when we calculate the accuracy score we get a very high score however our model is useless imagine telling a person that he has no flu when he actually has it is generally better in such cases to classify the person having the flu when he doesn't than to miss a case where someone does have the flu but the model says they don't two more formulas appear here recall and precision their difference is in the denominator in our problem terms Precision has false positives in the denominator meaning a high Precision value will mean the model does not predict flu when the person is healthy recall on the other hand targets false negatives meaning if we have very high recall the model identifies everyone who has the flu and does not miss anyone who is sick there is another measure that tries to maximize both recall and precision called F1 score it takes the harmonic mean of precision and recall meaning that you can get a high F1 score in the case you have high precision and recall thus we use recall precision and F1 score for imbalanced data sets while accuracy for balanced ones for multiclass scenarios the method is slightly different we will refer to it later there are also other important metrics such as Au and Roc curves for regression and unsupervised tasks the metrics are different they are more complex ones and we will talk about all those in the upcoming videos so stay with us if you want to learn more about artificial intelligence subscribe to our channel to be aware of the new videos press the like button and let's discuss AI in the comments section