Project 5.1

Data Processing:

persons = pd.read_csv("./data/persons.csv")
check_nan = persons['age'].isnull().values.any()
persons.dropna(inplace = True)
persons = persons.reset_index()


persons['age'] = persons['age'].astype(int)
persons['edu'] = persons['edu'].astype(int)

X = np.array(persons.drop(["wealthC", "wealthI"], axis = 1))
y = persons.wealthC
y1 = persons.wealthI

WealthC as the target variable

Linear Regression:

Unstandardized features:

Training: 0.73597
Testing: 0.73517
MSE Training: 0.44256
MSE Testing: 0.44356

Standardized features:

Training: 0.73596
Testing: 0.73517
MSE Training: 0.44259
MSE Testing: 0.44356

Here we can see that standardizing the features the set did not have any meaningful impact on the training or testing R^2 results of the linear regression

Ridge Regression:

Optimal alpha value: 2.22222222
Training score for this value: 0.736
Testing score for this value: 0.735

Lasso Regression:

Optimal alpha value: 0.000277
Training score for this value: 0.736
Testing score for this value: 0.735

WealthI as the target variable

Linear Regression:

Unstandardized features:

Training: 0.82596
Testing: 0.82513
MSE Training: 1748904156.35407
MSE Testing: 1753514701.91972

Standardized Features:

Training: 0.82594
Testing: 0.82512
MSE Training: 1749096642.20336
MSE Testing: 1753585858.54931

Ridge Regression:

Optimal alpha value: 142.632
Training score for this value: 0.826
Testing score for this value: 0.825

Lasso Regression:

Optimal alpha value: 1.500
Training score for this value: 0.826
Testing score for this value: 0.825

Results:

Which of the models produced the best results in predicting wealth of all persons throughout the smaller West African country being described?

For me, the two best models were the ridge and lasso regressions for the wealthI target variable in terms of R^2 scores with 0.825 testing values for both. This was followed closely by the linear regressions for the wealthI target as well. Because the wealthI models consistently outperform the wealthC target models, I would assume that the wealthI data is described much better by the features.

For both targets, I had almost no difference between the lasso and ridge regressions, and no meaningful differences between the standardized and non-standardized linear regressions.