How to preform multiple linear regression on a dataset in python with scikit-learn?

Question

My python code originally outputted these results, a list of dictionaries of the population of a census tract (basically an area of land) and the various land cover types. Here that is:

[{'Total Population:': 4585, 'Total Water Ice Cover': 2.848142234497044, 'Total Developed': 17.205368316575324, 'Total Barren Land': 0.22439908514219134, 'Total Forest': 34.40642126612868},

 {'Total Population:': 4751, 'Total Water Ice Cover': 1.047783534830167, 'Total Developed': 37.27115716753022, 'Total Barren Land': 0.11514104778353484, 'Total Forest': 19.11341393206678},

 {'Total Population:': 3214, 'Total Water Ice Cover': 0.09166603009701321, 'Total Developed': 23.50469788404247, 'Total Barren Land': 0.2597204186082041, 'Total Forest': 20.418608204109695},

 {'Total Population:': 5005, 'Total Water Ice Cover': 0.0, 'Total Developed': 66.37545713124746, 'Total Barren Land': 0.0, 'Total Forest': 10.68671271840715},

...
]

Then taking that code placing it into a pandas object:

import pandas as pd
df = pd.DataFrame(output)
print(df)
#   Total Barren Land  Total Developed  Total Forest  Total Population:  Total Water Ice Cover
#0           0.224399        17.205368     34.406421               4585               2.848142 
#1           0.115141        37.271157     19.113414               4751               1.047784 
#2           0.259720        23.504698     20.418608               3214               0.091666   
#3           0.000000        66.375457     10.686713               5005               1.047784 

Then to get the pearson 'r' correlation:

pd.set_option("precision",4)  # only show 4 digits

# remove 'Total ' from column names to make printing smaller
df.rename(columns=lambda x: x.replace("Total ", ""), inplace=True)  

corr = df.corr(method="pearson")
print(corr)
#                 Barren Land  Developed  Forest  Population:  Water Ice Cover
#Barren Land           1.0000    -0.9579  0.7361      -0.7772           0.4001
#Developed            -0.9579     1.0000 -0.8693       0.5736          -0.6194
#Forest                0.7361    -0.8693  1.0000      -0.1575           0.9114
#Population:          -0.7772     0.5736 -0.1575       1.0000           0.2612
#Water Ice Cover       0.4001    -0.6194  0.9114       0.2612           1.0000

Now I have all the pearson 'r' correlation values between population and various land cover types.

What I want to do now is calculate the multiple linear regression. I am trying to perform multiple linear regression between the population density and area percentage of the following surface covers and calculate the R2 of the regression: developed, class planted/cultivated class and maybe some other. Could this also be done through pandas?

Thank you

Answer 1

You can do a multiple regression with either Scikit-learn or Statsmodels.

You can see an exemple of multiple regression using scikit_learn here: Multiple linear regression in Python

As for Statsmodels, you can do something like that:

import statsmodels.api as sm    

X = df[[“variable_1”, “variable_2”]]
y = df[“target”]

model = sm.OLS(y, X).fit()
predictions = model.predict(X)
model.summary()