In our last blog we outlined the basics of supervised machine learning including the existence of labels. When possible, using labels in the learning process makes understanding the results of the learning easier and provides the user with a better experience.
Accuracy can be measured against ground truth, something not possible in unsupervised learning. Since accuracy can be measured, the algorithms themselves can be designed to maximize various competing aspects of accuracy, such as precision and recall or false positive and false negative rates. Not only can accuracy be measured, confidence in the accuracy of any predictions can also be quantified.
Example of labels being applied in Supervised Machine Learning
The model answers a clearly formulated question, typically “How should this sample be labeled, given the set of labels used for the training instances?” or “For this sample, what is the value of a given unmeasured feature?” There is no doubt as to what the results mean; the learner is simply trying to model the behavior of the subject matter experts in labeling unknown samples.
Looking for the comprehensive guide for understanding machine learning? Download the recently updated Guide for Machine Learning for Cybersecurity.