Reference:

5 Tribes of Machine Learning

THE GOAL OF MACHINE LEARNING

With machine learning, we want to program our machines to use labeled data to solve a given problem by learning how to extract patterns and relationships.

The goal of the Machine Learning model: It must detect relationships between many given patterns → Acquiring Knowledge from Information:

A simple example: Predicting House Prices - Train an ML Model to predict house prices based on provided features and values.

Each house in the dataset has multiple features (patterns), such as: Square footage, Number of bedrooms, Number of bathrooms, Location, Age, whether it has a pool or garage, etc.

The model analyzes relationships between these features and the price of the house. For instance, it may learn:

• Bigger houses usually cost more (relationship between square footage and price).
• Pools may add significant value in some regions but not others.
• Houses in certain zip codes are more expensive (relationship between location and price).
• Newer houses may have a higher value (relationship between age and price).

Once trained, the model can predict the price of a new, unseen house by detecting and applying the learned relationships between the input patterns.

<aside> 💡

If the ML model simply memorizes the data → Popularly known as Overfitting, it won’t be able to detect general patterns or relationships.

</aside>

⇒ We train a ML Model for it to make prediction, not to play memory game. The ML Model must generalize the data, not memorize it. If a model is just remembering the training set, it's likely overfitting, meaning it performs well on data it has seen but poorly on new, unseen one. This contradicts the purpose of machine learning, which is to make accurate predictions on new inputs. It does not need to make 100% right prediction, but at least 80%-99%.