OnlineNewsPopularity Data

This dataset has been widely used for research and education purposes, particularly in the field of machine learning and data analysis, to explore how different factors influence the popularity of online news articles. However, as it's been some time since my last update, I recommend checking the UCI Machine Learning Repository or other reliable sources for the most up-to-date information on the "OnlineNewsPopularity" dataset.

Summary

In this task, we predict the value of the shares column by pre-processing the data and checking different machine-learning models.

Model results with R2_score and training with linear regression without any preprocessing: 0.013436911487878178

Model results with R2_score and training with polynomial regression by preprocessing: 0.42482892634462005

Checking correlation

It can be seen that most of the features are not related to the target.

Therefore, it is not possible to remove a feature easily, because it is possible that some features have a non-linear relationship with the target, and also because the coefficients are very close to each other.

So, removing features in relation to their relationship with the target is not a good thing.

Checking the distribution of shares

Remove outlier records

IQR :

Due to the deletion of a large part of the data, we abandon this method.

Z-test :

Due to the non-normality of the distribution, we refrain from using this method.

Percentile:

By removing one percentage of outlier data :‌

Scaling

• Quantile Transformer

• power transformer

  

• -min-max-

  

Again scaling

Now we rescale the data that we scaled with the Quantile Transformer method

• min-mix

  

A very important point is that all data are normally between 0 and 1.

• power transformer

  

• RobustScaler

  

• StandardScaler

  

using min-max after Quantile

Linear model

R2_score : 0.15420224054686105

Polynomial model

R2_score : 0.42482892634462005