This project involves using R programming language on python runtime in google colab by using rmagic. The reason for not using R runtime is the unavailability of an authentic way to mount google drive on colab with R runtime, which poses an issue of being not able to use project specific datasets uploaded online.
%load_ext rpy2.ipythonThis will allow the python runtime of google colab to execute code written in R. Keep in mind that by using rmagic we will have to include %%R at the start of each cell.
The very first step in reproducing this code is to download the 'csv' files named 'TrainingDatas.csv', 'TD.csv', 'Target.csv' and upload them to your google drive. Once uploaded, mount your google drive with the following lines of code:
from pydrive.auth import GoogleAuth
from pydrive.drive import GoogleDrive
from google.colab import auth
from oauth2client.client import GoogleCredentials
auth.authenticate_user()
gauth = GoogleAuth()
gauth.credentials = GoogleCredentials.get_application_default()
drive = GoogleDrive(gauth)And then:
from google.colab import drive
drive.mount('/content/drive')Make a dedicated folder for this project which stores all the csv files along with the colab notebook. (In my case the drive folder is named as ML Project)
The dataset belongs to an insurance company trying to figure out the kind of customers interested in buying their new insursnce policy 'CARAVAN Insurance Policy'. The dataset is compiled by Dutch Data Mining Compnay Sentinal Machine Research. The dataset consists of 86 variables where 1-43 are socio-demographic information about the customers and 43-86 shows ownership of other insurance policies held by customers. Total observations are 9822 where 5822 (almost 60%) will be used for training and 4000 (almost 41%) will be used for testing. Machine learning algorithm, Decision Trees will be applied on the dataset to determine the kind of customers interested in buying the 'CARAVAN' insurance policy. Other techniques like Random Forests, boosted trees, Bagging Trees are no doubt more robust techniques but the objective of this project calls for decision trees as I have to build a 'customer profile' based on prior data to determine who will buy the said insurance policy. The variable names are keywords corresponding to the actual value that column holds. For exmaple the first column 'MOSTYPE' shows the customer subtype. The values of the column are numeric which encodes to the customer subtypes given at the end of this explaination. Hence you have to look at the data dictionary that I have created and listed below:
1 MOSTYPE Customer Subtype see L0
2 MAANTHUI Number of houses 1 – 10
3 MGEMOMV Avg size household 1 – 6
4 MGEMLEEF Avg age see L1
5 MOSHOOFD Customer main type see L2
6 MGODRK Roman catholic see L3
7 MGODPR Protestant ...
8 MGODOV Other religion
9 MGODGE No religion
10 MRELGE Married
11 MRELSA Living together
12 MRELOV Other relation
13 MFALLEEN Singles
14 MFGEKIND Household without children
15 MFWEKIND Household with children
16 MOPLHOOG High level education
17 MOPLMIDD Medium level education
18 MOPLLAAG Lower level education
19 MBERHOOG High status
20 MBERZELF Entrepreneur
21 MBERBOER Farmer
22 MBERMIDD Middle management
23 MBERARBG Skilled labourers
24 MBERARBO Unskilled labourers
25 MSKA Social class A
26 MSKB1 Social class B1
27 MSKB2 Social class B2
28 MSKC Social class C
29 MSKD Social class D
30 MHHUUR Rented house
31 MHKOOP Home owners
32 MAUT1 1 car
33 MAUT2 2 cars
34 MAUT0 No car
35 MZFONDS National Health Service
36 MZPART Private health insurance
37 MINKM30 Income < 30.000
38 MINK3045 Income 30-45.000
39 MINK4575 Income 45-75.000
40 MINK7512 Income 75-122.000
41 MINK123M Income >123.000
42 MINKGEM Average income
43 MKOOPKLA Purchasing power class
44 PWAPART Contribution private third party insurance see L4
45 PWABEDR Contribution third party insurance (firms) ...
46 PWALAND Contribution third party insurane (agriculture)
47 PPERSAUT Contribution car policies
48 PBESAUT Contribution delivery van policies
49 PMOTSCO Contribution motorcycle/scooter policies
50 PVRAAUT Contribution lorry policies
51 PAANHANG Contribution trailer policies
52 PTRACTOR Contribution tractor policies
53 PWERKT Contribution agricultural machines policies
54 PBROM Contribution moped policies
55 PLEVEN Contribution life insurances
56 PPERSONG Contribution private accident insurance policies
57 PGEZONG Contribution family accidents insurance policies
58 PWAOREG Contribution disability insurance policies
59 PBRAND Contribution fire policies
60 PZEILPL Contribution surfboard policies
61 PPLEZIER Contribution boat policies
62 PFIETS Contribution bicycle policies
63 PINBOED Contribution property insurance policies
64 PBYSTAND Contribution social security insurance policies
65 AWAPART Number of private third party insurance 1 - 12
66 AWABEDR Number of third party insurance (firms) ...
67 AWALAND Number of third party insurane (agriculture)
68 APERSAUT Number of car policies
69 ABESAUT Number of delivery van policies
70 AMOTSCO Number of motorcycle/scooter policies
71 AVRAAUT Number of lorry policies
72 AAANHANG Number of trailer policies
73 ATRACTOR Number of tractor policies
74 AWERKT Number of agricultural machines policies
75 ABROM Number of moped policies
76 ALEVEN Number of life insurances
77 APERSONG Number of private accident insurance policies
78 AGEZONG Number of family accidents insurance policies
79 AWAOREG Number of disability insurance policies
80 ABRAND Number of fire policies
81 AZEILPL Number of surfboard policies
82 APLEZIER Number of boat policies
83 AFIETS Number of bicycle policies
84 AINBOED Number of property insurance policies
85 ABYSTAND Number of social security insurance policies
86 CARAVAN Number of mobile home policies 0 - 1
The averaged variables having keywords L0, L1, L2, L3, L4 are listed below:
L0
1 High Income, expensive child
2 Very Important Provincials
3 High status seniors
4 Affluent senior apartments
5 Mixed seniors
6 Career and childcare
7 Dinki's (double income no kids)
8 Middle class families
9 Modern, complete families
10 Stable family
11 Family starters
12 Affluent young families
13 Young all american family
14 Junior cosmopolitan
15 Senior cosmopolitans
16 Students in apartments
17 Fresh masters in the city
18 Single youth
19 Suburban youth
20 Etnically diverse
21 Young urban have-nots
22 Mixed apartment dwellers
23 Young and rising
24 Young, low educated
25 Young seniors in the city
26 Own home elderly
27 Seniors in apartments
28 Residential elderly
29 Porchless seniors: no front yard
30 Religious elderly singles
31 Low income catholics
32 Mixed seniors
33 Lower class large families
34 Large family, employed child
35 Village families
36 Couples with teens 'Married with children'
37 Mixed small town dwellers
38 Traditional families
39 Large religous families
40 Large family farms
41 Mixed rurals
L1
1 20-30 years
2 30-40 years
3 40-50 years
4 50-60 years
5 60-70 years
6 70-80 years
L2
1 Successful hedonists
2 Driven Growers
3 Average Family
4 Career Loners
5 Living well
6 Cruising Seniors
7 Retired and Religeous
8 Family with grown ups
9 Conservative families
10 Farmers
L3
0 0%
1 1 - 10%
2 11 - 23%
3 24 - 36%
4 37 - 49%
5 50 - 62%
6 63 - 75%
7 76 - 88%
8 89 - 99%
9 100%
L4
0 f 0
1 f 1 – 49
2 f 50 – 99
3 f 100 – 199
4 f 200 – 499
5 f 500 – 999
6 f 1000 – 4999
7 f 5000 – 9999
8 f 10.000 - 19.999
9 f 20.000
Loading the datasets uploaded in the ML project folder by using the link of each file:
%%R
train_data<- read.csv('/content/drive/MyDrive/ML Project/TrainingDatas.csv')
test_data<- read.csv('/content/drive/MyDrive/ML Project/TD.csv')
target<- read.csv('/content/drive/MyDrive/ML Project/Target.csv')
target<-as.vector(target$CARAVAN)This section includes a rudamentary data analysis with the help of pie charts and graphs. Before starting any data science project, it is imperative to fish out important details about the individual variables along with their effect on the dependent variable. All important relational visualizations are done in the colab notebook with sufficent details. Look at the 'Exploratory Analysis' section of colab to know more.
The first step of data pre-processing includes checking for multi-colinearity among the data variables. It is important to remove highly collinear variables so that any unwanted bias can be removed from the output.
%%R
corr_mat<-cor(full_data)
temp<- which(abs(corr_mat) > 0.80, arr.ind=TRUE)
corr_df<-as.data.frame(temp)
col_corr_df<-corr_df$col
var<-colnames(full_data[col_corr_df])
new_corr_mat<-cbind(temp,var)
new_corr_df<-as.data.frame(new_corr_mat)
corr_filter<-new_corr_df$row == new_corr_df$col
final_corr_df<-new_corr_df[!corr_filter,]
names(final_corr_df)<-c("Variable_1_Index", "Variable_2_Index", "Variable 2")
final_corr_df| Variable 1 | Variable_1_Index | Variable_2_Index | Variable 2 |
|---|---|---|---|
| MOSHOOFD | 5 | 1 | MOSTYPE.1 |
| MOSTYPE.1 | 1 | 5 | MOSHOOFD |
| MRELOV | 12 | 10 | MRELGE.1 |
| MRELGE.1 | 10 | 12 | MRELOV |
| MHKOOP | 31 | 30 | MHHUUR.1 |
| MHHUUR.1 | 30 | 31 | MHKOOP |
| MZPART | 36 | 35 | MZFONDS.1 |
| MZFONDS.1 | 35 | 36 | MZPART |
| AWAPART | 65 | 44 | PWAPART.1 |
| AWABEDR | 66 | 45 | PWABEDR.1 |
| AWALAND | 67 | 46 | PWALAND.1 |
| APERSAUT | 68 | 47 | PPERSAUT.1 |
| ABESAUT | 69 | 48 | PBESAUT.1 |
| AMOTSCO | 70 | 49 | PMOTSCO.1 |
| AVRAAUT | 71 | 50 | PVRAAUT.1 |
| AAANHANG | 72 | 51 | PAANHANG.1 |
| ATRACTOR | 73 | 52 | PTRACTOR.1 |
| AWERKT | 74 | 53 | PWERKT.1 |
| ABROM | 75 | 54 | PBROM.1 |
| ALEVEN | 76 | 55 | PLEVEN.1 |
| APERSONG | 77 | 56 | PPERSONG.1 |
| AGEZONG | 78 | 57 | PGEZONG.1 |
| AWAOREG | 79 | 58 | PWAOREG.1 |
| ABRAND | 80 | 59 | PBRAND.1 |
| AZEILPL | 81 | 60 | PZEILPL.1 |
| APLEZIER | 82 | 61 | PPLEZIER.1 |
| AFIETS | 83 | 62 | PFIETS.1 |
| AINBOED | 84 | 63 | PINBOED.1 |
| ABYSTAND | 85 | 64 | PBYSTAND.1 |
| PWAPART.1 | 44 | 65 | AWAPART |
| PWABEDR.1 | 45 | 66 | AWABEDR |
| PWALAND.1 | 46 | 67 | AWALAND |
| PPERSAUT.1 | 47 | 68 | APERSAUT |
| PBESAUT.1 | 48 | 69 | ABESAUT |
| PMOTSCO.1 | 49 | 70 | AMOTSCO |
| PVRAAUT.1 | 50 | 71 | AVRAAUT |
| PAANHANG.1 | 51 | 72 | AAANHANG |
| PTRACTOR.1 | 52 | 73 | ATRACTOR |
| PWERKT.1 | 53 | 74 | AWERKT |
| PBROM.1 | 54 | 75 | ABROM |
| PLEVEN.1 | 55 | 76 | ALEVEN |
| PPERSONG.1 | 56 | 77 | APERSONG |
| PGEZONG.1 | 57 | 78 | AGEZONG |
| PWAOREG.1 | 58 | 79 | AWAOREG |
| PBRAND.1 | 59 | 80 | ABRAND |
| PZEILPL.1 | 60 | 81 | AZEILPL |
| PPLEZIER.1 | 61 | 82 | APLEZIER |
| PFIETS.1 | 62 | 83 | AFIETS |
| PINBOED.1 | 63 | 84 | AINBOED |
| PBYSTAND.1 | 64 | 85 | ABYSTAND |
The output of final_corr_df is a dataframe shwoing all persistent correlations between variables of the dataset (shown above). Each variable name is written against its index and the name of the variable it is most closely related with. For example; in first row the first variable MOSHOOFD is collinear with MOSTYPE, and so on.
Looking at the above correlation matrix, it is evident that we got two complete sets of variable to use under the column names, Variable 1 and Variable 2. We can either take the left hand side variables, Variable 1 or the right hand side variables, Variable 2 to create a model. Hence we have to remove the redundant variables from the dataset:
%%R
rm_var<-c(5,10,30,35,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85)
new_train_data <- train_data [, -rm_var]Hence the number of features have been reduced to 61.
Now its time to randomize the data to exclude any danger of data bias if any:
%%R
random<-sample(1:nrow(new_train_data))
new_train_data<- new_train_data[random,]For the purpose of cutomer profiling, the algorithm of decision trees is used. The function rpart() of R library rpart is used to train the model on our dataset of 61 variables.
%%R
model<- rpart(formula=CARAVAN ~ .,data= new_train_data)
modelNow that the model is trained and stored in the variable `model', it is time to make some predictions. Before moving on to predict our target dataset, there is a need to prune our model. Pruning is necessary because it avoids data redundancy in the output decision trees hence making it more robust. If a tree is too large with many redundant nodes, it performs poorly on the new data. Furthermore, a large tree is susceptible to overfitting and hence generalizes very poorly.
%%R
print(model$cptable)
cp<-model$cptable[which.min(model$cptable[,"xerror"]),"CP"]
model_cp<-prune(tree =model,cp=cp)model_cp is our new, pruned model. The predictions will be made by utilizing this model:
pred<-predict(object = model_cp, new_test_data)
pred<-as.vector(pred)
for (i in 1: 4000){
if (pred[i]<0.2){
pred[i]=0
}
else{
pred[i]=1
}
}To evaluate the performance of the model there is a need to develop metrics against which it is judged. A good perforamance metric is a confusion matrix which is often used for classification tasks.
%%R
actual<-target
predicted<-pred
cm<-confusionMatrix(actual, predicted)
print(cm)| Confusion Matrix | True | False |
|---|---|---|
| True | 3479 | 191 |
| False | 283 | 47 |
As shown in above confusion matrix the model was successful in predicting the customer choice of either purchasing or not purchasing CARAVAN insurance policy by 88.1%. The misclassification error turns out to be 11%.
%%R
prediction <- factor(as.factor(pred), c(0, 1), labels = c("Not purchased", "Purchased"))
target <- factor(as.factor(target), c(0, 1), labels = c("Not purchased", "Purchased"))
tb<-table(prediction)
print(tb)
tb_t<-table(target)
print(tb_t)| Title | Predicted | Actual |
|---|---|---|
| Purchased | 330 | 238 |
| Not Purchased | 3670 | 3762 |
As seen from above table, the model misclassified 92 observations of 'Not purchased' as 'Purchased'.
The rmse of the model turns out to be:
rms<-rmse(target,pred)
print(rms)0.3442383
For in depth analysis and customer profiling see Analysis Section of the google colab.