# -*- coding: utf-8 -*-

"""

Created on Sat Dec 5 21:36:10 2020

@author: Lenovo

"""

import pandas as pd

import matplotlib.pyplot as plt

Fraud_Data = pd.read_csv("C:\Users\Lenovo\Documents\Assignments\Fraud_check.csv")

Fraud_Data.head()

Fraud_Data.isnull().any()#to check if we have null valuesin dataset or not

#so there are no null values in the dataset

Fraud_Data.dtypes# to check the data types

Fraud_Data.describe()### to check the summary of the dataset

Fraud_Data.info()

Fraud_Data['Undergrad'],class_names = pd.factorize(Fraud_Data['Undergrad'])

Fraud_Data.Undergrad

print(class_names)

Fraud_Data['Marital.Status'],class_names2=pd.factorize(Fraud_Data['Marital.Status'])

Fraud_Data['Marital.Status']

print(class_names2)

Fraud_Data['Urban'],class_names3=pd.factorize(Fraud_Data['Urban'])

Fraud_Data['Urban']

print(class_names3)

Fraud_Data.info()

Fraud_Data['Taxable.Income'].plot.hist()

plt.show()

Fraud_Data['Taxable.Income'].describe()

#count 600.000000

#mean 55208.375000

#std 26204.827597

#min 10003.000000

#25% 32871.500000

#50% 55074.500000

#75% 78611.750000

#max 99619.000000

#Converting the Sales column which is continuous into categorical

category = pd.cut(Fraud_Data['Taxable.Income'],bins=[0,30000,99619],labels=['Risky','Good'])

Fraud_Data.insert(0,'Taxable_Group',category)

Fraud_Data

import seaborn as sns

sns.pairplot(Fraud_Data)

Fraud_Data['Taxable_Group'].unique()

Fraud_Data.Taxable_Group.value_counts()#Good 476

#Risky 124

######### Features Selection ##########

colnames = list(Fraud_Data.columns)

colnames

predictors = colnames[1:]#excluding 1st column all

predictors #['Undergrad',

#'Marital.Status',

#'Taxable.Income',

#'City.Population',

#'Work.Experience',

#'Urban']feature variables

target = colnames[0]#only 1st column

target #'Taxable_Group' target variable

########## Splitting data into training and testing data set #############

import numpy as np

# np.random.uniform(start,stop,size) will generate array of real numbers with size = size

#### another way to split data into train and test

from sklearn.model_selection import train_test_split

train,test = train_test_split(Fraud_Data,test_size = 0.2)

test

test.shape # (120, 7)

train

train.shape #(480, 7)

############## Decision Tree Model building ###############

from sklearn.tree import DecisionTreeClassifier

help(DecisionTreeClassifier)

model = DecisionTreeClassifier(criterion = 'entropy')

model.fit(train[predictors],train[target])

preds = model.predict(test[predictors])

preds

from sklearn import metrics #Import scikit-learn metrics module for accuracy calculation

print("Accuracy:",metrics.accuracy_score(test[target], preds))

#Accuracy: 0.9916666666666667

pd.Series(preds).value_counts()

#moderate 43

#Good 93

#Risky 27

#dtype: int64

pd.crosstab(test[target],preds)

#col_0 Good Risky

#Taxable_Group

#Risky 0 26

#Good 93 1

# Accuracy = train

np.mean(train.Taxable_Group == model.predict(train[predictors]))#1.0

# Accuracy = Test

np.mean(preds==test.Taxable_Group) # 0.9916666666666667

model.score(test[predictors],test[target])

from sklearn.tree import export_graphviz

from sklearn.externals.six import StringIO #:pip install --upgrade scikit-learn==0.23.1

from IPython.display import Image

import pydotplus

import io

dot_data = StringIO()

export_graphviz(model, out_file=dot_data,

filled=True, rounded=True,

special_characters=True,feature_names = predictors,class_names=['Good','Risky'])

graph = pydotplus.graph_from_dot_data(dot_data.getvalue())

graph.write_png('Fraud.png')

Image(graph.create_png())