Detect fake profiles in online social networks using Random
Forest
In [54]: import sys
import csv
import datetime
import numpy as np
import pandas as pd
import [Link] as plt
from datetime import datetime
import [Link] as gender
from [Link] import Imputer
from sklearn import cross_validation
from sklearn import metrics
from sklearn import preprocessing
from [Link] import roc_curve, auc
from [Link] import RandomForestClassifier
from sklearn.cross_validation import StratifiedKFold, train_test_split
from sklearn.grid_search import GridSearchCV
from [Link] import accuracy_score
from sklearn.learning_curve import learning_curve
from [Link] import classification_report
from [Link] import confusion_matrix
%matplotlib inline
function for reading dataset from csv files
In [55]: def read_datasets():
""" Reads users profile from csv files """
genuine_users = pd.read_csv("data/[Link]")
fake_users = pd.read_csv("data/[Link]")
# print genuine_users.columns
# print genuine_users.describe()
#print fake_users.describe()
x=[Link]([genuine_users,fake_users])
y=len(fake_users)*[0] + len(genuine_users)*[1]
return x,y
function for predicting sex using name of person
In [56]: def predict_sex(name):
sex_predictor = [Link](unknown_value=u"unknown",case_sensitiv
e=False)
first_name= [Link](' ').[Link](0)
sex= first_name.apply(sex_predictor.get_gender)
sex_dict={'female': -2, 'mostly_female': -1,'unknown':0,'mostly_mal
e':1, 'male': 2}
sex_code = [Link](sex_dict).astype(int)
return sex_code
function for feature engineering
In [57]: def extract_features(x):
lang_list = list(enumerate([Link](x['lang'])))
lang_dict = { name : i for i, name in lang_list }
[Link][:,'lang_code'] = x['lang'].map( lambda x: lang_dict[x]).astype(i
nt)
[Link][:,'sex_code']=predict_sex(x['name'])
feature_columns_to_use = ['statuses_count','followers_count','friend
s_count','favourites_count','listed_count','sex_code','lang_code']
x=[Link][:,feature_columns_to_use]
return x
function for ploting learning curve
In [60]: def plot_learning_curve(estimator, title, X, y, ylim=None, cv=None,
n_jobs=1, train_sizes=[Link](.1, 1.0, 5)):
[Link]()
[Link](title)
if ylim is not None:
[Link](*ylim)
[Link]("Training examples")
[Link]("Score")
train_sizes, train_scores, test_scores = learning_curve(
estimator, X, y, cv=cv, n_jobs=n_jobs, train_sizes=train_sizes)
train_scores_mean = [Link](train_scores, axis=1)
train_scores_std = [Link](train_scores, axis=1)
test_scores_mean = [Link](test_scores, axis=1)
test_scores_std = [Link](test_scores, axis=1)
[Link]()
plt.fill_between(train_sizes, train_scores_mean - train_scores_std,
train_scores_mean + train_scores_std, alpha=0.1,
color="r")
plt.fill_between(train_sizes, test_scores_mean - test_scores_std,
test_scores_mean + test_scores_std, alpha=0.1, colo
r="g")
[Link](train_sizes, train_scores_mean, 'o-', color="r",
label="Training score")
[Link](train_sizes, test_scores_mean, 'o-', color="g",
label="Cross-validation score")
[Link](loc="best")
return plt
function for plotting confusion matrix
In [61]: def plot_confusion_matrix(cm, title='Confusion matrix', cmap=[Link]
s):
target_names=['Fake','Genuine']
[Link](cm, interpolation='nearest', cmap=cmap)
[Link](title)
[Link]()
tick_marks = [Link](len(target_names))
[Link](tick_marks, target_names, rotation=45)
[Link](tick_marks, target_names)
plt.tight_layout()
[Link]('True label')
[Link]('Predicted label')
function for plotting ROC curve
In [62]: def plot_roc_curve(y_test, y_pred):
false_positive_rate, true_positive_rate, thresholds = roc_curve(y_tes
t, y_pred)
print "False Positive rate: ",false_positive_rate
print "True Positive rate: ",true_positive_rate
roc_auc = auc(false_positive_rate, true_positive_rate)
[Link]('Receiver Operating Characteristic')
[Link](false_positive_rate, true_positive_rate, 'b',
label='AUC = %0.2f'% roc_auc)
[Link](loc='lower right')
[Link]([0,1],[0,1],'r--')
[Link]([-0.1,1.2])
[Link]([-0.1,1.2])
[Link]('True Positive Rate')
[Link]('False Positive Rate')
[Link]()
Function for training data using Random Forest
In [63]: def train(X_train,y_train,X_test):
""" Trains and predicts dataset with a Random Forest classifier """
clf=RandomForestClassifier(n_estimators=40,oob_score=True)
[Link](X_train,y_train)
print("The best classifier is: ",clf)
# Estimate score
scores = cross_validation.cross_val_score(clf, X_train,y_train, cv=5)
print scores
print('Estimated score: %0.5f (+/- %0.5f)' % ([Link](), [Link]
d() / 2))
title = 'Learning Curves (Random Forest)'
plot_learning_curve(clf, title, X_train, y_train, cv=5)
[Link]()
# Predict
y_pred = [Link](X_test)
return y_test,y_pred
In [64]: print "reading datasets.....\n"
x,y=read_datasets()
[Link]()
reading datasets.....
Out[64]: id statuses_count followers_count friends_count favourites_count
count 2.818000e+03 2818.000000 2818.000000 2818.000000 2818.000000
mean 5.374889e+08 1672.198368 371.105039 395.363023 234.541164
std 2.977005e+08 4884.669157 8022.631339 465.694322 1445.847248
min 3.610511e+06 0.000000 0.000000 0.000000 0.000000
25% 3.620867e+08 35.000000 17.000000 168.000000 0.000000
50% 6.162253e+08 77.000000 26.000000 306.000000 0.000000
75% 6.177673e+08 1087.750000 111.000000 519.000000 37.000000
max 1.391998e+09 79876.000000 408372.000000 12773.000000 44349.000000
In [65]: print "extracting featues.....\n"
x=extract_features(x)
print [Link]
print [Link]()
extracting featues.....
Index([u'statuses_count', u'followers_count', u'friends_count',
u'favourites_count', u'listed_count', u'sex_code', u'lang_code'],
dtype='object')
statuses_count followers_count friends_count favourites_count \
count 2818.000000 2818.000000 2818.000000 2818.000000
mean 1672.198368 371.105039 395.363023 234.541164
std 4884.669157 8022.631339 465.694322 1445.847248
min 0.000000 0.000000 0.000000 0.000000
25% 35.000000 17.000000 168.000000 0.000000
50% 77.000000 26.000000 306.000000 0.000000
75% 1087.750000 111.000000 519.000000 37.000000
max 79876.000000 408372.000000 12773.000000 44349.000000
listed_count sex_code lang_code
count 2818.000000 2818.000000 2818.000000
mean 2.818666 -0.180270 2.851313
std 23.480430 1.679125 1.992950
min 0.000000 -2.000000 0.000000
25% 0.000000 -2.000000 1.000000
50% 0.000000 0.000000 1.000000
75% 1.000000 2.000000 5.000000
max 744.000000 2.000000 7.000000
In [66]: print "spliting datasets in train and test dataset...\n"
X_train,X_test,y_train,y_test = train_test_split(x, y, test_size=0.20, ran
dom_state=44)
spliting datasets in train and test dataset...
In [67]: print "training datasets.......\n"
y_test,y_pred = train(X_train,y_train,X_test)
training datasets.......
('The best classifier is: ', RandomForestClassifier(bootstrap=True, clas
s_weight=None, criterion='gini',
max_depth=None, max_features='auto', max_leaf_nodes=None,
min_samples_leaf=1, min_samples_split=2,
min_weight_fraction_leaf=0.0, n_estimators=40, n_jobs=1,
oob_score=True, random_state=None, verbose=0, warm_start=Fals
e))
[ 0.93791574 0.93791574 0.94678492 0.9578714 0.93777778]
Estimated score: 0.94365 (+/- 0.00395)
In [68]: print 'Classification Accuracy on Test dataset: ' ,accuracy_score(y_test,
y_pred)
Classification Accuracy on Test dataset: 0.941489361702
In [70]: cm=confusion_matrix(y_test, y_pred)
print('Confusion matrix, without normalization')
print(cm)
plot_confusion_matrix(cm)
Confusion matrix, without normalization
[[265 3]
[ 30 266]]
In [71]: cm_normalized = [Link]('float') / [Link](axis=1)[:, [Link]]
print('Normalized confusion matrix')
print(cm_normalized)
plot_confusion_matrix(cm_normalized, title='Normalized confusion matrix')
Normalized confusion matrix
[[ 0.98880597 0.01119403]
[ 0.10135135 0.89864865]]
In [72]: print(classification_report(y_test, y_pred, target_names=['Fake','Genuin
e']))
precision recall f1-score support
Fake 0.90 0.99 0.94 268
Genuine 0.99 0.90 0.94 296
avg / total 0.95 0.94 0.94 564
In [73]: plot_roc_curve(y_test, y_pred)
False Positive rate: [ 0. 0.01119403 1. ]
True Positive rate: [ 0. 0.89864865 1. ]
