Feature Scaling and Machine Learning

University:
Stanford University
Course:
CS229 | Machine Learning
Academic year:

2023
Views:

281

Pages:

1
Author:

anon

Feature Scaling and Machine Learning: Its Importance The variety and distribution of input features can have a significant impact on machine learning algorithms. For this reason, it's crucial to scale the features that go into a machine learning model correctly. Assuring that each feature contributes roughly the same amount to the outcome and preventing features with greater ranges from predominating the model are both benefits of feature scaling. What does Feature Scaling mean? The method of feature scaling involves changing the input characteristics so that they have equivalent value ranges. It is a crucial step in the machine learning stage of pre-processing data. Assuring that each feature contributes roughly the same amount to the outcome and preventing features with greater ranges from predominating the model are both benefits of feature scaling. Feature Scaling Techniques Scaling features can be done in a number of ways, such as by dividing by the maximum, mean normalizing, and Z-score normalization. Divide by the highest number Simply multiply each original feature value by the range's highest value to scale features by dividing by the maximum. For instance, if feature 𝑥1 has a range of 3 to 2,000, one technique to scale 𝑥1 is to divide each original value of 𝑥1 by the maximum value in the range, which in this case is 2,000. The scale 𝑥1 that results will go from 0.15 to 1. Normalization to Mean The features are rescaled to be centered around zero during mean normalization. Find the average (mean) of the feature on your training set before normalizing it. Next, divide each feature value by the difference between the maximum and minimum values after deducting the mean from each value. For instance, if the average of feature 𝑥1 is 600, then each value for 𝑥1 is subtracted 600, and the difference between 2,000 and 300, where 2,000 is the maximum and 300 is the minimum, is divided by 600. The range of the mean normalized 𝑥1 as a consequence is 0.18 to 0.82. Normalization of Z-Score It is necessary to figure out each feature's standard deviation for Z-score normalization. Subtract the mean from each feature value and divide by the standard deviation to normalize a feature's Z-score. For instance, to obtain the Z-score normalized x 1 if feature x 1 has a standard deviation of 450 and a mean of 600, subtract 600 from each value of x 1 and divide by 450.