crosshyou

主にクロス表(分割表)分析をしようかなと思いはじめましたが、あまりクロス表の分析はできず。R言語の練習ブログになっています。

OECD Discriminatory family code data analysis 8 - Comparing some classification methods.

Photo by Marek Piwnicki on Unsplash 

www.crosshyou.info

This post is following of above post.

In this post, I will do some classification methods.

Firstly, I make binary variable.

I made a binary variable named high, that shows 1 when lpc_gdp is hgher than the mdeian and 0 when lower than median.

Before making models, I load caret package/

Then, I make LPM: Linear Probability Model, which is liear regression model with OLS estimate.

"atwm": Attitides Towards Working Mother is only significant variable.

Then, make a prediction and calculating how much prediction is correct.

Accuracy is 0.7941. This means 79.4% is correctly predicted.

Next, I use SVM: Support Vector Machine. I use e1071 package svm() function.

with SVM model, what is accuracy?

Accuracy is 0.8529. So SVM is betther than LPM.

Next, I use GAM: Generalized Additive Model. I use gam() function in mgcv package.

In GAM, s(atwm) and s(l_inf) are significant variables.

Let's plot GAM model.

The plot shows s(em) is not needed s().
So, let' make another GAM model, this time s(l_unem).

plot GAM2

Let's make predictions with GAM model

Let's get accuracy og GAM.

Accuracy us 0.9118.

How about GAM2 model?

Accuracy is 0.8824. So, the fist GAM model is better.

Next, I use tree model.

I use tree() function in tree package. In this model, "atwm" is the most important variable and l_inf is next. Others are not important.

Let's get accuracy of tree model.

Accuracy is 0.8235.

Next, I use k-NN model.

I use knn3() function in caret package.

Let's get accuracy of knn.

Accuracy is 0.7353.

Then, let's compare those methods accuracy.

GAM model has the highest accuracy.
That's it. Thank you!

 

To read the 1st post

 

www.crosshyou.info