Apply logistic regression to categorize whether a county had high mortality rate due to contamination

In [55]:

    

import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
from sklearn.linear_model import LogisticRegression


    

In [24]:

    

df = pd.read_csv('hanford.csv')


    

In [25]:

    

df.describe()


    

    
Out[25]:






  

    

      

      
Exposure
      
Mortality
    
  
  

    

      
count
      
9.000000
      
9.000000
    
    

      
mean
      
4.617778
      
157.344444
    
    

      
std
      
3.491192
      
34.791346
    
    

      
min
      
1.250000
      
113.500000
    
    

      
25%
      
2.490000
      
130.100000
    
    

      
50%
      
3.410000
      
147.100000
    
    

      
75%
      
6.410000
      
177.900000
    
    

      
max
      
11.640000
      
210.300000
    
  

In [26]:

    

iqr = df.quantile(q=0.75) - df.quantile(q=0.25)
iqr


    

    
Out[26]:





Exposure      3.92
Mortality    47.80
dtype: float64

In [27]:

    

ual = df.quantile(q=0.75) + (iqr * 1.5)
ual


    

    
Out[27]:





Exposure      12.29
Mortality    249.60
dtype: float64

In [28]:

    

lal = df.quantile(q=0.25) - (iqr * 1.5)
lal


    

    
Out[28]:





Exposure     -3.39
Mortality    58.40
dtype: float64

In [29]:

    

df.plot(kind='scatter', x='Exposure', y='Mortality')


    

    
Out[29]:





<matplotlib.axes._subplots.AxesSubplot at 0x11003d080>






    

In [30]:

    

for value in df['Exposure']:
    if value < ual['Exposure']:
        print(value)


    

    




2.49
2.57
3.41
1.25
1.62
3.83
11.64
6.41
8.34

In [31]:

    

# Find new reasonable threshold!
# Choosing 6


    

In [32]:

    

df['high_exposure'] = df['Exposure'].apply(lambda x:1 if x>6 else 0)


    

In [33]:

    

df


    

    
Out[33]:






  

    

      

      
County
      
Exposure
      
Mortality
      
high_exposure
    
  
  

    

      
0
      
Umatilla
      
2.49
      
147.1
      
0
    
    

      
1
      
Morrow
      
2.57
      
130.1
      
0
    
    

      
2
      
Gilliam
      
3.41
      
129.9
      
0
    
    

      
3
      
Sherman
      
1.25
      
113.5
      
0
    
    

      
4
      
Wasco
      
1.62
      
137.5
      
0
    
    

      
5
      
HoodRiver
      
3.83
      
162.3
      
0
    
    

      
6
      
Portland
      
11.64
      
207.5
      
1
    
    

      
7
      
Columbia
      
6.41
      
177.9
      
1
    
    

      
8
      
Clatsop
      
8.34
      
210.3
      
1
    
  

In [34]:

    

# dataset = df[['Mortality']].join([pd.get_dummies(df['Exposure'],prefix="Exposure"),df.high_exposure])


    

In [35]:

    

# dataset


    

In [36]:

    

from sklearn.linear_model import LogisticRegression


    

In [37]:

    

lm = LogisticRegression()


    

In [43]:

    

x = np.asarray(df[['Mortality']])
y = np.asarray(df['high_exposure'])


    

In [46]:

    

lm = lm.fit(x,y)


    

In [50]:

    

lm.score(x,y)


    

    
Out[50]:





0.66666666666666663

In [47]:

    

lm.coef_


    

    
Out[47]:





array([[ 0.00259684]])

In [48]:

    

lm.intercept_


    

    
Out[48]:





array([-0.74701648])

In [58]:

    

plt.plot(x,lm.coef_[0]*x+lm.intercept_[0])


    

    
Out[58]:





[<matplotlib.lines.Line2D at 0x110283710>]






    

In [60]:

    

df['high_mortality'] = df['Mortality'].apply(lambda x:1 if x>150 else 0)


    

In [61]:

    

lm2 = LogisticRegression()


    

In [65]:

    

x2 = np.asarray(df[['Exposure']])
y2 = np.asarray(df['high_mortality'])


    

In [67]:

    

lm2 = lm2.fit(x2,y2)


    

In [68]:

    

lm2.predict(50)


    

    
Out[68]:





array([1])

In [69]:

    

# According to the prediction the mortality rate is high at an exposure level of 50.


    

In [ ]: