You are working as a machine learning specialist for a car rental firm that wishes to use machine learning to optimize the cost per mile of its rental cars based on geographic region.
You have a large car rental database for use in your model training that has features such as rental car type, rental geographic region, miles driven, average regional gas price, etc.
During your exploratory data analysis tasks, you notice that your feature seat contains outliers for the miles driven and average regional gas price.
These outliers are likely to make your regression algorithm-based model unstable.
What data preparation technique can you use to prepare your data for your model training?
Click on the arrows to vote for the correct answer
A. B. C. D.Correct Answer: D.
Option A is incorrect.
Both normalization and standardization are types of data scaling that machine learning specialists use to prepare their data for training and inference.
Normalization rescales your data so that all values range from 0 to 1
Therefore, normalization doesn't handle outliers well.
Option B is incorrect.
Quantile Binning is used to categorize feature values into bins.
This technique would not reduce the effect of your outliers since the outliers would skew whichever bin in which they are placed.
Option C is incorrect.
The Min-Max Scaling term is another name for normalizing your data.
Normalization doesn't handle outliers as well as standardization.
Option D is correct.
Standardization centers your feature values around the mean.
So, it has no bounding range.
Therefore, standardization handles outliers much better than normalization.
References:
Please see the article titled Feature Scaling for Machine Learning: Understanding the Difference Between Normalization vs.
Standardization (https://www.analyticsvidhya.com/blog/2020/04/feature-scaling-machine-learning-normalization-standardization/),
The Amazon Machine Learning developer guide titled Data Transformations Reference (https://docs.aws.amazon.com/machine-learning/latest/dg/data-transformations-reference.html)
Outliers are data points that are significantly different from other data points in the same feature. They can be caused by a variety of reasons such as data entry errors, measurement errors, or genuine variations in the data. Outliers can have a significant impact on the performance of machine learning algorithms, especially regression-based models, as they can skew the distribution of the data and negatively impact the model's ability to generalize to new data.
In this scenario, the car rental firm wants to use machine learning to optimize the cost per mile of its rental cars based on geographic region. The data available for model training includes features such as rental car type, rental geographic region, miles driven, average regional gas price, etc. During the exploratory data analysis tasks, it was found that the feature "miles driven" and "average regional gas price" contain outliers. This can make the regression algorithm-based model unstable.
To prepare the data for model training, one can use the following data preparation techniques:
A. Normalize your features to reduce the effect of the outlier data: Normalization is a process of scaling numerical data to a range between 0 and 1. This technique can be used to reduce the impact of outliers by scaling the data to a common range, which helps to make the data more uniform. However, this technique does not remove outliers from the data.
B. Use Quantile Binning of your features to reduce the effect of the outlier data: Quantile binning is a technique that involves dividing the data into equal-sized bins based on their percentile ranks. This technique can be used to reduce the impact of outliers by replacing them with the upper or lower limit of the bin in which they fall. This technique can be useful if the outliers are due to measurement errors or data entry errors, but may not be effective if the outliers are genuine variations in the data.
C. Min-Max Scale your features to reduce the effect of the outlier data: Min-Max scaling is a technique that involves scaling numerical data to a range between a minimum and maximum value. This technique can be used to reduce the impact of outliers by scaling the data to a common range, which helps to make the data more uniform. However, like normalization, this technique does not remove outliers from the data.
D. Standardize your features to reduce the effect of the outlier data: Standardization is a technique that involves scaling numerical data to have zero mean and unit variance. This technique can be used to reduce the impact of outliers by scaling the data in a way that makes it more resistant to outliers. Standardization is more effective than normalization or Min-Max scaling for removing the effect of outliers from the data.
In summary, the best data preparation technique to prepare the data for model training, given that the feature "miles driven" and "average regional gas price" contain outliers, is to standardize the features to reduce the effect of the outlier data.