Crime per population estimator

This notebook is trying to estimate the violent crime rate per population based on data from https://archive.ics.uci.edu/ml/datasets/Communities+and+Crime using different linear regression models on multivariate data.

It is based on examples from sklearn and is licensed under Apache 2.0 open source license.

In [1]:
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt

import sklearn.metrics
import sklearn.linear_model
import sklearn.preprocessing
import sklearn.pipeline

%matplotlib inline
In [2]:
# Communities and Crime Dataset 
# https://archive.ics.uci.edu/ml/datasets/Communities+and+Crime
df_crime_dataset = pd.read_csv('communities.data.txt', na_values='?', header=None)
# this is just the headers from the community.names file
headers_df = pd.read_csv('data_headers.txt', sep=' ', header=None)
df_crime_dataset.columns = headers_df[1].tolist()
#df_crime_dataset
In [3]:
df_crime_dataset.shape
Out[3]:
(1994, 128)
In [18]:
df_crime_dataset.describe()
Out[18]:
state county community fold population householdsize racepctblack racePctWhite racePctAsian racePctHisp ... LandArea PopDens PctUsePubTrans PolicCars PolicOperBudg LemasPctPolicOnPatr LemasGangUnitDeploy LemasPctOfficDrugUn PolicBudgPerPop ViolentCrimesPerPop
count 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 ... 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000 1994.000000
mean 28.683551 24.191575 18924.709629 5.493982 0.057593 0.463395 0.179629 0.753716 0.153681 0.144022 ... 0.065231 0.232854 0.161685 0.026093 0.012272 0.111760 0.070461 0.094052 0.031209 0.237979
std 16.397553 86.058017 27897.731489 2.873694 0.126906 0.163717 0.253442 0.244039 0.208877 0.232492 ... 0.109459 0.203092 0.229055 0.104581 0.062672 0.270038 0.228819 0.240328 0.097190 0.232985
min 1.000000 0.000000 0.000000 1.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 ... 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
25% 12.000000 0.000000 0.000000 3.000000 0.010000 0.350000 0.020000 0.630000 0.040000 0.010000 ... 0.020000 0.100000 0.020000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.070000
50% 34.000000 0.000000 0.000000 5.000000 0.020000 0.440000 0.060000 0.850000 0.070000 0.040000 ... 0.040000 0.170000 0.070000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.150000
75% 42.000000 17.000000 39596.250000 8.000000 0.050000 0.540000 0.230000 0.940000 0.170000 0.160000 ... 0.070000 0.280000 0.190000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.330000
max 56.000000 840.000000 94597.000000 10.000000 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000 ... 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000 1.000000

8 rows × 127 columns

In [5]:
# set nan to zero
df_crime_dataset.fillna(value=0, inplace=True)
# shuffling
df_crime_dataset = df_crime_dataset.sample(frac=1).reset_index(drop=True)
In [6]:
sns.pairplot(df_crime_dataset, 
             x_vars=['medIncome','FemalePctDiv','PolicPerPop'], 
             y_vars='ViolentCrimesPerPop', size=7, aspect=0.7)
Out[6]:
<seaborn.axisgrid.PairGrid at 0x219a04c5a58>
In [7]:
sns.pairplot(df_crime_dataset, 
             x_vars=['PctHousNoPhone','NumInShelters','PopDens'], 
             y_vars='ViolentCrimesPerPop', size=7, aspect=0.7)
Out[7]:
<seaborn.axisgrid.PairGrid at 0x219a3f18978>
In [8]:
# create arrays of features X and labels y (crime rate)
np_feature_set = df_crime_dataset.iloc[:,5:-1].as_matrix()
np_label_set = df_crime_dataset.iloc[:,-1].as_matrix()
In [9]:
# Split data in train set and test set
n_samples = np_feature_set.shape[0]
cutoff_train = int(n_samples * 0.9)
x_train, y_train = np_feature_set[:cutoff_train], np_label_set[:cutoff_train]
x_test, y_test = np_feature_set[cutoff_train:], np_label_set[cutoff_train:]
In [17]:
# training

alpha = 0.0001

def my_trainer(model, x_train, y_train, x_test, y_test):
    y_pred = model.fit(x_train, y_train).predict(x_test)
    r2_score = sklearn.metrics.r2_score(y_test, y_pred)
    print(model)
    print("r^2 on test data : %f \n" % r2_score)

# Lasso
lasso = sklearn.linear_model.Lasso(alpha=alpha)
my_trainer(lasso, x_train, y_train, x_test, y_test)

# ElasticNet
elnet = sklearn.linear_model.ElasticNet(alpha=alpha, l1_ratio=0.7)
my_trainer(elnet, x_train, y_train, x_test, y_test)

#LassoLars
lasso_lars = sklearn.linear_model.LassoLars(alpha=alpha)
my_trainer(lasso_lars, x_train, y_train, x_test, y_test)

#Polynominal Preposessing with ElNet
#poly_prep = sklearn.preprocessing.PolynomialFeatures(degree=3)
#poly_elnet_model = sklearn.pipeline.Pipeline([('poly', poly_prep),
#                  ('elnet', elnet)])
#my_trainer(poly_elnet_model, x_train, y_train, x_test, y_test)

Lasso(alpha=0.0001, copy_X=True, fit_intercept=True, max_iter=1000,
   normalize=False, positive=False, precompute=False, random_state=None,
   selection='cyclic', tol=0.0001, warm_start=False)
r^2 on test data : 0.626746 

ElasticNet(alpha=0.0001, copy_X=True, fit_intercept=True, l1_ratio=0.7,
      max_iter=1000, normalize=False, positive=False, precompute=False,
      random_state=None, selection='cyclic', tol=0.0001, warm_start=False)
r^2 on test data : 0.631559 

LassoLars(alpha=0.0001, copy_X=True, eps=2.2204460492503131e-16,
     fit_intercept=True, fit_path=True, max_iter=500, normalize=True,
     positive=False, precompute='auto', verbose=False)
r^2 on test data : 0.623220
In [14]:
plt.plot(elnet.coef_, color='lightgreen', linewidth=2,
         label='Elastic net coefficients')
plt.plot(lasso.coef_, color='gold', linewidth=2,
         label='Lasso coefficients')
plt.plot(lasso_lars.coef_, color='red', linewidth=2,
         label='Lasso Lars coefficients')

plt.legend(loc='best')
plt.show()
In [ ]: