What Is Supervised Learning? Breaking It Down for Everyone
Learn briefly about what supervised learning is and how it is actually conducted during training
As the name “Supervised” implies being watched over, machine learning in this process is managed by training with labeled data, where each piece of information is tagged with the correct answer.
Supervised learning is conducted at a foundational level in five steps: Data, Model, Training, Evaluating, and Inference.
Data Collection
The first and foremost step is to collect data. It should be large in size and high in diversity. Why? Because the more diverse the data is, the better the model can generalize, ensuring better prediction results in the real world.
The data could be in the form of images, prices, weather information, etc., based on the task types of the target label: Classification and Regression.
Datasets contain both Features and Labels. Here, features are the information that we provide, which the model counts as values.
The label, on the other hand, is the answer we want to know. So, what a model does is take into account all of the features to calculate and give its prediction.
For instance, if the label prediction is for the rainfall amount in the months of June, July, then it takes all the relevant historical data of the past few decades from the feature values (such as temperature, humidity, cloud coverage, and so on) and gives its prediction of the estimated amount of rainfall.
Therefore, data quality is very important. In the tech world, this is known as the “Garbage In, Garbage Out” rule, which means your final predictions are only as good as your data.
A model trained with flawed or biased data can only provide ineffective predictions, no matter how good your model is.
Read More: What Is Machine Learning? Breaking It Down for Everyone
Model and Training
The model is basically the mathematical formula or the process of how the features are to be calculated and transformed into a prediction.
When data is collected and refined, it is then given to a model for training. Upon receiving the data, the model makes an initial guess, calculates errors against the true labels, and adjusts its internal settings.
In this way, a model gradually learns the correct relationship between features and labels to produce a trustworthy outcome.
Evaluating and Inference
Finally, when the training period is completed, after millions, or in some cases even billions, of individual calculations to learn properly, a model goes through its evaluation in a testing phase.
An independent, unseen labeled dataset is used for testing the performance to ensure the model can handle new data accurately.
If the results are not satisfactory, more training and evaluating follow before deploying the model for real-world application.
Once the model is installed for practical use to predict new unlabeled data, it is called inference.
Editorial Note:
The views and insights in this article are the author’s own original write-up, developed through research and study of Google’s Machine Learning course, AI tools, and Google Search.
Reference
# https://developers.google.com/machine-learning/intro-to-ml/supervised