Overview of the three steps

1. Data Pre-Processing - Importing, cleaning the data and splitting into training or test sets
2. Modelling - Build, train the model and make predictions
3. Evaluation - Calculate performance metrics and make a verdict

Data Pre-Processing

Splitting into Training and Test sets

• Let’s say we have a model that predicts the sale prices of cars by assessing the age of the car and its odometer usage
• From the 20 cars, 20% of the data (4 cars) will be the test set and the 80% will be used to build the model
• The resultant model will then be applied to the 20% test set, in order to predict certain prices and because it was separated in advance, we can compare with the actual values.

Feature Scaling

• Feature scaling is a way to normalize/standardize variables in order to find better similarities in data
• In this example, both blue/purple or red/purple can be considered similar, unless it’s feature scaled
• Feature scaling is applied to columns, NOT to data inside a row