{"id":1673,"date":"2021-12-28T10:23:54","date_gmt":"2021-12-28T03:23:54","guid":{"rendered":"https:\/\/mintea.blog\/?p=1673"},"modified":"2021-12-28T11:23:44","modified_gmt":"2021-12-28T04:23:44","slug":"1673","status":"publish","type":"post","link":"https:\/\/mintea.blog\/?p=1673","title":{"rendered":"Predicting CLV (demo Python)"},"content":{"rendered":"

Predicting CLV (demo Python)<\/strong><\/p>\n

Customer Lifetime Value prediction<\/p>\n

The problem<\/p>\n

In this notebook we look at the data we got via this\u00a0Kaggle dataset (CreditCard_dataset)<\/a>. It involves the car insurance customer lifetime value.<\/p>\n

Customer Lifetime Value Prediction( CLV ) value refers to net profit attributed to the entire future relationship with a customer. A bank will use different predictive analytic approaches to predict the revenue that can be generated from any customer in the future. This helps the banks in segmentating the customers in specific groups based on their CLV.<\/p>\n

Identifying customers with high future values will enable the organization to keep maintaining good relationships with such customers. It can be done by investing more time and resources on them such as better prices, offers, discounts, customer care services, etc.<\/p>\n

Finding and engaging reliable and profitable customers has always been a great challenge for banks. With the increasing competition, the banks need to keep a check on each and every activity of their customers for utilizing their resources effectively.<\/p>\n

To solve this problem, Data Science in banking is being used for extracting actionable insights concerning customer behaviors and expectations. Using Data Science models for predicting the CLV of a customer will help a bank to take some suitable decisions for their growth and profit.<\/p>\n

\"CLV\"<\/p>\n

Import the important libraries \/ packages<\/p>\n

These packages are needed to load and use the dataset<\/p>\n

In\u00a0[1]:<\/strong><\/p>\n

\n
import<\/span> pandas<\/span> as<\/span> pd<\/span> #we use this to load, read and transform the dataset<\/span>\r\nimport<\/span> numpy<\/span> as<\/span> np<\/span> #we use this for statistical analysis<\/span>\r\nimport<\/span> matplotlib.pyplot<\/span> as<\/span> plt<\/span> #we use this to visualize the dataset<\/span>\r\nimport<\/span> seaborn<\/span> as<\/span> sns<\/span> #we use this to make countplots<\/span>\r\nimport<\/span> sklearn.metrics<\/span> as<\/span> sklm<\/span> #This is to test the models<\/span>\r\n<\/pre>\n<\/div>\n
Load and explore the dataset<\/p>\n
The data is all in one csv file. In this next step I will first load the data to see how this looks like<\/p>\n
In\u00a0[2]:<\/strong><\/p>\n
\n
#here we load the data<\/span>\r\ndata =<\/span> pd.<\/span>read_csv('\/kaggle\/input\/credit-card-data\/Fn-UseC_-Marketing-Customer-Value-Analysis.csv'<\/span>)\r\n#and immediately I would like to see how this dataset looks like<\/span>\r\ndata.<\/span>head()\r\n<\/pre>\n<\/div>\n
Out[2]:<\/strong><\/p>\n
\n\n\n\n\n\n\n\n\n\n
<\/th>\nCustomer<\/th>\nState<\/th>\nCustomer Lifetime Value<\/th>\nResponse<\/th>\nCoverage<\/th>\nEducation<\/th>\nEffective To Date<\/th>\nEmploymentStatus<\/th>\nGender<\/th>\nIncome<\/th>\n…<\/th>\nMonths Since Policy Inception<\/th>\nNumber of Open Complaints<\/th>\nNumber of Policies<\/th>\nPolicy Type<\/th>\nPolicy<\/th>\nRenew Offer Type<\/th>\nSales Channel<\/th>\nTotal Claim Amount<\/th>\nVehicle Class<\/th>\nVehicle Size<\/th>\n<\/tr>\n<\/thead>\n
0<\/td>\nBU79786<\/td>\nWashington<\/td>\n2763.519279<\/td>\nNo<\/td>\nBasic<\/td>\nBachelor<\/td>\n2\/24\/11<\/td>\nEmployed<\/td>\nF<\/td>\n56274<\/td>\n…<\/td>\n5<\/td>\n0<\/td>\n1<\/td>\nCorporate Auto<\/td>\nCorporate L3<\/td>\nOffer1<\/td>\nAgent<\/td>\n384.811147<\/td>\nTwo-Door Car<\/td>\nMedsize<\/td>\n<\/tr>\n
1<\/td>\nQZ44356<\/td>\nArizona<\/td>\n6979.535903<\/td>\nNo<\/td>\nExtended<\/td>\nBachelor<\/td>\n1\/31\/11<\/td>\nUnemployed<\/td>\nF<\/td>\n0<\/td>\n…<\/td>\n42<\/td>\n0<\/td>\n8<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer3<\/td>\nAgent<\/td>\n1131.464935<\/td>\nFour-Door Car<\/td>\nMedsize<\/td>\n<\/tr>\n
2<\/td>\nAI49188<\/td>\nNevada<\/td>\n12887.431650<\/td>\nNo<\/td>\nPremium<\/td>\nBachelor<\/td>\n2\/19\/11<\/td>\nEmployed<\/td>\nF<\/td>\n48767<\/td>\n…<\/td>\n38<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer1<\/td>\nAgent<\/td>\n566.472247<\/td>\nTwo-Door Car<\/td>\nMedsize<\/td>\n<\/tr>\n
3<\/td>\nWW63253<\/td>\nCalifornia<\/td>\n7645.861827<\/td>\nNo<\/td>\nBasic<\/td>\nBachelor<\/td>\n1\/20\/11<\/td>\nUnemployed<\/td>\nM<\/td>\n0<\/td>\n…<\/td>\n65<\/td>\n0<\/td>\n7<\/td>\nCorporate Auto<\/td>\nCorporate L2<\/td>\nOffer1<\/td>\nCall Center<\/td>\n529.881344<\/td>\nSUV<\/td>\nMedsize<\/td>\n<\/tr>\n
4<\/td>\nHB64268<\/td>\nWashington<\/td>\n2813.692575<\/td>\nNo<\/td>\nBasic<\/td>\nBachelor<\/td>\n2\/3\/11<\/td>\nEmployed<\/td>\nM<\/td>\n43836<\/td>\n…<\/td>\n44<\/td>\n0<\/td>\n1<\/td>\nPersonal Auto<\/td>\nPersonal L1<\/td>\nOffer1<\/td>\nAgent<\/td>\n138.130879<\/td>\nFour-Door Car<\/td>\nMedsize<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
5 rows \u00d7 24 columns<\/p>\n
In\u00a0[3]:<\/strong><\/p>\n
#now let’s look closer at the dataset we got<\/em><\/p>\n
<\/a> data.info()<\/p>\n
\"Text<\/p>\n
It seems that we have a lot of text \/ category information (these are of the Dtype ‘object’) and a few numerical columns (Dtypes ‘int64’ and ‘float64’).<\/p>\n
The column ‘Customer Lifetime Value’ is the column we would like to predict.<\/p>\n
In\u00a0[4]:<\/strong><\/p>\n
<\/a> data.shape<\/p>\n
Out[4]:<\/strong><\/p>\n
(9134, 24)<\/p>\n
The dataset consists of 9134 rows and 24 columns.<\/p>\n
In\u00a0[5]:<\/strong><\/p>\n
<\/a> data.describe()<\/p>\n
Out[5]:<\/strong><\/p>\n
\n\n\n\n\n\n\n\n\n\n\n\n\n
<\/th>\nCustomer Lifetime Value<\/th>\nIncome<\/th>\nMonthly Premium Auto<\/th>\nMonths Since Last Claim<\/th>\nMonths Since Policy Inception<\/th>\nNumber of Open Complaints<\/th>\nNumber of Policies<\/th>\nTotal Claim Amount<\/th>\n<\/tr>\n<\/thead>\n
count<\/td>\n9134.000000<\/td>\n9134.000000<\/td>\n9134.000000<\/td>\n9134.000000<\/td>\n9134.000000<\/td>\n9134.000000<\/td>\n9134.000000<\/td>\n9134.000000<\/td>\n<\/tr>\n
mean<\/td>\n8004.940475<\/td>\n37657.380009<\/td>\n93.219291<\/td>\n15.097000<\/td>\n48.064594<\/td>\n0.384388<\/td>\n2.966170<\/td>\n434.088794<\/td>\n<\/tr>\n
std<\/td>\n6870.967608<\/td>\n30379.904734<\/td>\n34.407967<\/td>\n10.073257<\/td>\n27.905991<\/td>\n0.910384<\/td>\n2.390182<\/td>\n290.500092<\/td>\n<\/tr>\n
min<\/td>\n1898.007675<\/td>\n0.000000<\/td>\n61.000000<\/td>\n0.000000<\/td>\n0.000000<\/td>\n0.000000<\/td>\n1.000000<\/td>\n0.099007<\/td>\n<\/tr>\n
25%<\/td>\n3994.251794<\/td>\n0.000000<\/td>\n68.000000<\/td>\n6.000000<\/td>\n24.000000<\/td>\n0.000000<\/td>\n1.000000<\/td>\n272.258244<\/td>\n<\/tr>\n
50%<\/td>\n5780.182197<\/td>\n33889.500000<\/td>\n83.000000<\/td>\n14.000000<\/td>\n48.000000<\/td>\n0.000000<\/td>\n2.000000<\/td>\n383.945434<\/td>\n<\/tr>\n
75%<\/td>\n8962.167041<\/td>\n62320.000000<\/td>\n109.000000<\/td>\n23.000000<\/td>\n71.000000<\/td>\n0.000000<\/td>\n4.000000<\/td>\n547.514839<\/td>\n<\/tr>\n
max<\/td>\n83325.381190<\/td>\n99981.000000<\/td>\n298.000000<\/td>\n35.000000<\/td>\n99.000000<\/td>\n5.000000<\/td>\n9.000000<\/td>\n2893.239678<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
It seems that we have some strange outliers for the CLV and claim amounts. We will look and handle these later on.<\/p>\n
In\u00a0[6]:<\/strong><\/p>\n
<\/a> data.describe(include=’O’)<\/p>\n
Out[6]:<\/strong><\/p>\n
\n\n\n\n\n\n\n\n\n
<\/th>\nCustomer<\/th>\nState<\/th>\nResponse<\/th>\nCoverage<\/th>\nEducation<\/th>\nEffective To Date<\/th>\nEmploymentStatus<\/th>\nGender<\/th>\nLocation Code<\/th>\nMarital Status<\/th>\nPolicy Type<\/th>\nPolicy<\/th>\nRenew Offer Type<\/th>\nSales Channel<\/th>\nVehicle Class<\/th>\nVehicle Size<\/th>\n<\/tr>\n<\/thead>\n
count<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n9134<\/td>\n<\/tr>\n
unique<\/td>\n9134<\/td>\n5<\/td>\n2<\/td>\n3<\/td>\n5<\/td>\n59<\/td>\n5<\/td>\n2<\/td>\n3<\/td>\n3<\/td>\n3<\/td>\n9<\/td>\n4<\/td>\n4<\/td>\n6<\/td>\n3<\/td>\n<\/tr>\n
top<\/td>\nYD27780<\/td>\nCalifornia<\/td>\nNo<\/td>\nBasic<\/td>\nBachelor<\/td>\n1\/10\/11<\/td>\nEmployed<\/td>\nF<\/td>\nSuburban<\/td>\nMarried<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer1<\/td>\nAgent<\/td>\nFour-Door Car<\/td>\nMedsize<\/td>\n<\/tr>\n
freq<\/td>\n1<\/td>\n3150<\/td>\n7826<\/td>\n5568<\/td>\n2748<\/td>\n195<\/td>\n5698<\/td>\n4658<\/td>\n5779<\/td>\n5298<\/td>\n6788<\/td>\n3426<\/td>\n3752<\/td>\n3477<\/td>\n4621<\/td>\n6424<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
In\u00a0[7]:<\/strong><\/p>\n
\n
#Let's see what the options are in the text columns with two or three options (the objects)<\/span>\r\nprint<\/span>('Response: '<\/span>+<\/span> str<\/span>(data['Response'<\/span>].<\/span>unique()))\r\nprint<\/span>('Coverage: '<\/span>+<\/span> str<\/span>(data['Coverage'<\/span>].<\/span>unique()))\r\nprint<\/span>('Education: '<\/span>+<\/span> str<\/span>(data['Education'<\/span>].<\/span>unique()))\r\nprint<\/span>('Employment Status: '<\/span>+<\/span> str<\/span>(data['EmploymentStatus'<\/span>].<\/span>unique()))\r\nprint<\/span>('Gender: '<\/span> +<\/span> str<\/span>(data['Gender'<\/span>].<\/span>unique()))\r\nprint<\/span>('Location Code: '<\/span> +<\/span> str<\/span>(data['Location Code'<\/span>].<\/span>unique()))\r\nprint<\/span>('Married: '<\/span> +<\/span> str<\/span>(data['Marital Status'<\/span>].<\/span>unique()))\r\nprint<\/span>('Policy Type: '<\/span> +<\/span> str<\/span>(data['Policy Type'<\/span>].<\/span>unique()))\r\nprint<\/span>('Vehicle Size: '<\/span> +<\/span> str<\/span>(data['Vehicle Size'<\/span>].<\/span>unique()))\r\n<\/pre>\n<\/div>\n
\"Text<\/p>\n
Customer Lifetime Value<\/p>\n
As Customer Lifetime Value is the column we want to predict, let’s explore this column in the training dataset.<\/p>\n
The formula to calculate the CLV:<\/p>\n
\"CLV<\/p>\n
In\u00a0[8]:<\/strong><\/p>\n
\n
#As this is a numeric, thus continous number, I will use a scatterplot to see if there is a pattern.<\/span>\r\nplt.<\/span>hist(data['Customer Lifetime Value'<\/span>], bins =<\/span> 10<\/span>)\r\nplt.<\/span>title(\"Customer Lifetime Value\"<\/span>) #Assign title<\/span>\r\nplt.<\/span>xlabel(\"Value\"<\/span>) #Assign x label<\/span>\r\nplt.<\/span>ylabel(\"Customers\"<\/span>) #Assign y label<\/span>\r\nplt.<\/span>show()\r\n<\/pre>\n<\/div>\n
\"Chart,<\/p>\n
In\u00a0[9]:<\/strong><\/p>\n
<\/a> plt.boxplot(data[‘Customer Lifetime Value’])<\/p>\n
Out[9]:<\/strong><\/p>\n
\"Text<\/p>\n
\"A<\/p>\n
In\u00a0[10]:<\/strong><\/p>\n
#We see that there are some great outliers here.<\/em><\/p>\n
#let’s look closer to these outliers over 50000<\/em><\/p>\n
<\/a> outliers = data[data[‘Customer Lifetime Value’] > 50000]<\/p>\n
<\/a> outliers.head(25)<\/p>\n
Out[10]:<\/strong><\/p>\n
\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
<\/th>\nCustomer<\/th>\nState<\/th>\nCustomer Lifetime Value<\/th>\nResponse<\/th>\nCoverage<\/th>\nEducation<\/th>\nEffective To Date<\/th>\nEmploymentStatus<\/th>\nGender<\/th>\nIncome<\/th>\n…<\/th>\nMonths Since Policy Inception<\/th>\nNumber of Open Complaints<\/th>\nNumber of Policies<\/th>\nPolicy Type<\/th>\nPolicy<\/th>\nRenew Offer Type<\/th>\nSales Channel<\/th>\nTotal Claim Amount<\/th>\nVehicle Class<\/th>\nVehicle Size<\/th>\n<\/tr>\n<\/thead>\n
79<\/td>\nOM82309<\/td>\nCalifornia<\/td>\n58166.55351<\/td>\nNo<\/td>\nBasic<\/td>\nBachelor<\/td>\n2\/27\/11<\/td>\nEmployed<\/td>\nM<\/td>\n61321<\/td>\n…<\/td>\n30<\/td>\n1<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer2<\/td>\nBranch<\/td>\n427.631210<\/td>\nLuxury Car<\/td>\nSmall<\/td>\n<\/tr>\n
1974<\/td>\nYC54142<\/td>\nWashington<\/td>\n74228.51604<\/td>\nNo<\/td>\nExtended<\/td>\nHigh School or Below<\/td>\n1\/26\/11<\/td>\nUnemployed<\/td>\nM<\/td>\n0<\/td>\n…<\/td>\n34<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L1<\/td>\nOffer1<\/td>\nBranch<\/td>\n1742.400000<\/td>\nLuxury Car<\/td>\nMedsize<\/td>\n<\/tr>\n
2190<\/td>\nKI58952<\/td>\nCalifornia<\/td>\n51337.90677<\/td>\nNo<\/td>\nPremium<\/td>\nCollege<\/td>\n2\/24\/11<\/td>\nEmployed<\/td>\nF<\/td>\n72794<\/td>\n…<\/td>\n47<\/td>\n1<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L2<\/td>\nOffer1<\/td>\nWeb<\/td>\n50.454459<\/td>\nSUV<\/td>\nLarge<\/td>\n<\/tr>\n
2908<\/td>\nEN65835<\/td>\nArizona<\/td>\n58753.88046<\/td>\nNo<\/td>\nPremium<\/td>\nBachelor<\/td>\n1\/6\/11<\/td>\nEmployed<\/td>\nF<\/td>\n24964<\/td>\n…<\/td>\n84<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L2<\/td>\nOffer2<\/td>\nAgent<\/td>\n888.000000<\/td>\nSUV<\/td>\nMedsize<\/td>\n<\/tr>\n
3145<\/td>\nCL79250<\/td>\nNevada<\/td>\n52811.49112<\/td>\nNo<\/td>\nBasic<\/td>\nBachelor<\/td>\n1\/8\/11<\/td>\nUnemployed<\/td>\nM<\/td>\n0<\/td>\n…<\/td>\n70<\/td>\n0<\/td>\n2<\/td>\nCorporate Auto<\/td>\nCorporate L2<\/td>\nOffer2<\/td>\nAgent<\/td>\n873.600000<\/td>\nLuxury Car<\/td>\nSmall<\/td>\n<\/tr>\n
3760<\/td>\nAZ84403<\/td>\nOregon<\/td>\n61850.18803<\/td>\nNo<\/td>\nExtended<\/td>\nCollege<\/td>\n2\/4\/11<\/td>\nUnemployed<\/td>\nF<\/td>\n0<\/td>\n…<\/td>\n29<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L1<\/td>\nOffer3<\/td>\nBranch<\/td>\n1142.400000<\/td>\nLuxury SUV<\/td>\nMedsize<\/td>\n<\/tr>\n
4126<\/td>\nJT47995<\/td>\nArizona<\/td>\n60556.19213<\/td>\nNo<\/td>\nExtended<\/td>\nCollege<\/td>\n1\/1\/11<\/td>\nUnemployed<\/td>\nF<\/td>\n0<\/td>\n…<\/td>\n45<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer1<\/td>\nWeb<\/td>\n979.200000<\/td>\nLuxury SUV<\/td>\nLarge<\/td>\n<\/tr>\n
4915<\/td>\nDU50092<\/td>\nOregon<\/td>\n56675.93768<\/td>\nNo<\/td>\nPremium<\/td>\nCollege<\/td>\n1\/24\/11<\/td>\nEmployed<\/td>\nF<\/td>\n77237<\/td>\n…<\/td>\n93<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L1<\/td>\nOffer4<\/td>\nWeb<\/td>\n1358.400000<\/td>\nLuxury SUV<\/td>\nMedsize<\/td>\n<\/tr>\n
5279<\/td>\nSK66747<\/td>\nWashington<\/td>\n66025.75407<\/td>\nNo<\/td>\nBasic<\/td>\nBachelor<\/td>\n2\/22\/11<\/td>\nEmployed<\/td>\nM<\/td>\n33481<\/td>\n…<\/td>\n46<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer1<\/td>\nAgent<\/td>\n1194.892002<\/td>\nLuxury SUV<\/td>\nMedsize<\/td>\n<\/tr>\n
5716<\/td>\nFQ61281<\/td>\nOregon<\/td>\n83325.38119<\/td>\nNo<\/td>\nExtended<\/td>\nHigh School or Below<\/td>\n1\/31\/11<\/td>\nEmployed<\/td>\nM<\/td>\n58958<\/td>\n…<\/td>\n74<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer1<\/td>\nCall Center<\/td>\n1108.800000<\/td>\nLuxury Car<\/td>\nSmall<\/td>\n<\/tr>\n
6252<\/td>\nBP23267<\/td>\nCalifornia<\/td>\n73225.95652<\/td>\nNo<\/td>\nExtended<\/td>\nBachelor<\/td>\n2\/9\/11<\/td>\nEmployed<\/td>\nF<\/td>\n39547<\/td>\n…<\/td>\n21<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer1<\/td>\nBranch<\/td>\n969.600000<\/td>\nLuxury SUV<\/td>\nMedsize<\/td>\n<\/tr>\n
6461<\/td>\nOY68395<\/td>\nOregon<\/td>\n55277.44589<\/td>\nNo<\/td>\nBasic<\/td>\nHigh School or Below<\/td>\n1\/30\/11<\/td>\nEmployed<\/td>\nF<\/td>\n40740<\/td>\n…<\/td>\n60<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L2<\/td>\nOffer1<\/td>\nWeb<\/td>\n950.400000<\/td>\nLuxury SUV<\/td>\nLarge<\/td>\n<\/tr>\n
6554<\/td>\nAH58807<\/td>\nArizona<\/td>\n51426.24815<\/td>\nNo<\/td>\nBasic<\/td>\nCollege<\/td>\n1\/9\/11<\/td>\nEmployed<\/td>\nF<\/td>\n84650<\/td>\n…<\/td>\n39<\/td>\n3<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L2<\/td>\nOffer1<\/td>\nAgent<\/td>\n660.474274<\/td>\nLuxury Car<\/td>\nMedsize<\/td>\n<\/tr>\n
6569<\/td>\nLW64678<\/td>\nCalifornia<\/td>\n51016.06704<\/td>\nNo<\/td>\nPremium<\/td>\nMaster<\/td>\n2\/19\/11<\/td>\nEmployed<\/td>\nF<\/td>\n25167<\/td>\n…<\/td>\n76<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer2<\/td>\nAgent<\/td>\n422.494292<\/td>\nSUV<\/td>\nSmall<\/td>\n<\/tr>\n
6584<\/td>\nXF89906<\/td>\nArizona<\/td>\n58207.12842<\/td>\nNo<\/td>\nExtended<\/td>\nHigh School or Below<\/td>\n1\/13\/11<\/td>\nDisabled<\/td>\nM<\/td>\n29295<\/td>\n…<\/td>\n50<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer1<\/td>\nAgent<\/td>\n1328.839129<\/td>\nLuxury SUV<\/td>\nLarge<\/td>\n<\/tr>\n
7283<\/td>\nKH55886<\/td>\nOregon<\/td>\n67907.27050<\/td>\nNo<\/td>\nPremium<\/td>\nBachelor<\/td>\n2\/5\/11<\/td>\nEmployed<\/td>\nM<\/td>\n78310<\/td>\n…<\/td>\n18<\/td>\n1<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L1<\/td>\nOffer1<\/td>\nAgent<\/td>\n151.711475<\/td>\nSports Car<\/td>\nMedsize<\/td>\n<\/tr>\n
7303<\/td>\nFB95288<\/td>\nCalifornia<\/td>\n64618.75715<\/td>\nNo<\/td>\nExtended<\/td>\nHigh School or Below<\/td>\n1\/17\/11<\/td>\nUnemployed<\/td>\nM<\/td>\n0<\/td>\n…<\/td>\n40<\/td>\n1<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer1<\/td>\nBranch<\/td>\n1562.400000<\/td>\nLuxury Car<\/td>\nSmall<\/td>\n<\/tr>\n
7556<\/td>\nJZ23377<\/td>\nOregon<\/td>\n57520.50151<\/td>\nNo<\/td>\nPremium<\/td>\nCollege<\/td>\n1\/20\/11<\/td>\nEmployed<\/td>\nF<\/td>\n48367<\/td>\n…<\/td>\n34<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L3<\/td>\nOffer2<\/td>\nBranch<\/td>\n772.800000<\/td>\nSUV<\/td>\nMedsize<\/td>\n<\/tr>\n
7835<\/td>\nQT84069<\/td>\nOregon<\/td>\n50568.25912<\/td>\nNo<\/td>\nExtended<\/td>\nMaster<\/td>\n2\/28\/11<\/td>\nEmployed<\/td>\nM<\/td>\n82081<\/td>\n…<\/td>\n62<\/td>\n0<\/td>\n2<\/td>\nPersonal Auto<\/td>\nPersonal L1<\/td>\nOffer2<\/td>\nBranch<\/td>\n753.760098<\/td>\nLuxury SUV<\/td>\nSmall<\/td>\n<\/tr>\n
8825<\/td>\nUS30122<\/td>\nCalifornia<\/td>\n61134.68307<\/td>\nNo<\/td>\nBasic<\/td>\nCollege<\/td>\n2\/28\/11<\/td>\nUnemployed<\/td>\nM<\/td>\n0<\/td>\n…<\/td>\n75<\/td>\n0<\/td>\n2<\/td>\nCorporate Auto<\/td>\nCorporate L3<\/td>\nOffer2<\/td>\nBranch<\/td>\n2275.265075<\/td>\nLuxury Car<\/td>\nMedsize<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
20 rows \u00d7 24 columns<\/p>\n
In\u00a0[11]:<\/strong><\/p>\n
<\/a> outliers.info()<\/p>\n
\"Text<\/p>\n
Looks like there are only 20 rows of the 9134 rows that have a lifetime value of more than 50000. We will leave this as is for now<\/p>\n
Handling missing values<\/p>\n
Let’s continue with handling the missing values in this dataset. Let’s see where and how many missing values there are in this dataset.<\/p>\n
In\u00a0[12]:<\/strong><\/p>\n
#let’s look in what columns there are missing values<\/em><\/p>\n
<\/a> data.isnull().sum().sort_values(ascending = False)<\/p>\n
Out[12]:<\/strong><\/p>\n
\"Text<\/p>\n
There seem to be no missing values in this dataset.<\/p>\n
Making the text columns Numeric<\/p>\n
We first need to make all column input numeric to use them further on. This is what I will do now.<\/p>\n
In\u00a0[13]:<\/strong><\/p>\n
#First we drop the customer column, as this is a unique identifier and will bias the model<\/em><\/p>\n
<\/a> data = data.drop(labels = [‘Customer’], axis = 1)<\/p>\n
In\u00a0[14]:<\/strong><\/p>\n
#let’s load the required packages<\/em><\/p>\n
from sklearn.preprocessing import <\/a>LabelEncoder<\/p>\n
<\/a> le = LabelEncoder()<\/p>\n
In\u00a0[15]:<\/strong><\/p>\n
\n
# Let's transform the categorical variables to continous variables<\/span>\r\ncolumn_names =<\/span> ['Response'<\/span>, 'Coverage'<\/span>, 'Education'<\/span>, 'Effective To Date'<\/span>, 'EmploymentStatus'<\/span>,\r\n'Gender'<\/span>, 'Location Code'<\/span>, 'Marital Status'<\/span>,\r\n'Policy Type'<\/span>, 'Policy'<\/span>, 'Renew Offer Type'<\/span>,\r\n'Sales Channel'<\/span>, 'Vehicle Class'<\/span>, 'Vehicle Size'<\/span>, 'State'<\/span>]\r\nfor<\/span> col in<\/span> column_names:\r\n     data[col] =<\/span> le.<\/span>fit_transform(data[col])\r\ndata.<\/span>head()\r\n<\/pre>\n<\/div>\n
Out[15]:<\/strong><\/p>\n
\n\n\n\n\n\n\n\n\n\n
<\/th>\nState<\/th>\nCustomer Lifetime Value<\/th>\nResponse<\/th>\nCoverage<\/th>\nEducation<\/th>\nEffective To Date<\/th>\nEmploymentStatus<\/th>\nGender<\/th>\nIncome<\/th>\nLocation Code<\/th>\n…<\/th>\nMonths Since Policy Inception<\/th>\nNumber of Open Complaints<\/th>\nNumber of Policies<\/th>\nPolicy Type<\/th>\nPolicy<\/th>\nRenew Offer Type<\/th>\nSales Channel<\/th>\nTotal Claim Amount<\/th>\nVehicle Class<\/th>\nVehicle Size<\/th>\n<\/tr>\n<\/thead>\n
0<\/td>\n4<\/td>\n2763.519279<\/td>\n0<\/td>\n0<\/td>\n0<\/td>\n47<\/td>\n1<\/td>\n0<\/td>\n56274<\/td>\n1<\/td>\n…<\/td>\n5<\/td>\n0<\/td>\n1<\/td>\n0<\/td>\n2<\/td>\n0<\/td>\n0<\/td>\n384.811147<\/td>\n5<\/td>\n1<\/td>\n<\/tr>\n
1<\/td>\n0<\/td>\n6979.535903<\/td>\n0<\/td>\n1<\/td>\n0<\/td>\n24<\/td>\n4<\/td>\n0<\/td>\n0<\/td>\n1<\/td>\n…<\/td>\n42<\/td>\n0<\/td>\n8<\/td>\n1<\/td>\n5<\/td>\n2<\/td>\n0<\/td>\n1131.464935<\/td>\n0<\/td>\n1<\/td>\n<\/tr>\n
2<\/td>\n2<\/td>\n12887.431650<\/td>\n0<\/td>\n2<\/td>\n0<\/td>\n41<\/td>\n1<\/td>\n0<\/td>\n48767<\/td>\n1<\/td>\n…<\/td>\n38<\/td>\n0<\/td>\n2<\/td>\n1<\/td>\n5<\/td>\n0<\/td>\n0<\/td>\n566.472247<\/td>\n5<\/td>\n1<\/td>\n<\/tr>\n
3<\/td>\n1<\/td>\n7645.861827<\/td>\n0<\/td>\n0<\/td>\n0<\/td>\n12<\/td>\n4<\/td>\n1<\/td>\n0<\/td>\n1<\/td>\n…<\/td>\n65<\/td>\n0<\/td>\n7<\/td>\n0<\/td>\n1<\/td>\n0<\/td>\n2<\/td>\n529.881344<\/td>\n3<\/td>\n1<\/td>\n<\/tr>\n
4<\/td>\n4<\/td>\n2813.692575<\/td>\n0<\/td>\n0<\/td>\n0<\/td>\n52<\/td>\n1<\/td>\n1<\/td>\n43836<\/td>\n0<\/td>\n…<\/td>\n44<\/td>\n0<\/td>\n1<\/td>\n1<\/td>\n3<\/td>\n0<\/td>\n0<\/td>\n138.130879<\/td>\n0<\/td>\n1<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
5 rows \u00d7 23 columns<\/p>\n
In\u00a0[16]:<\/strong><\/p>\n
<\/a> data.dtypes<\/p>\n
Out[16]:<\/strong><\/p>\n
\"Text<\/p>\n
As my model can not handle floats, we will change these to integers.<\/p>\n
In\u00a0[17]:<\/strong><\/p>\n
\n
data['Customer Lifetime Value'<\/span>] =<\/span> data['Customer Lifetime Value'<\/span>].<\/span>astype(int<\/span>)\r\ndata['Total Claim Amount'<\/span>] =<\/span> data['Total Claim Amount'<\/span>].<\/span>astype(int<\/span>)\r\n<\/pre>\n<\/div>\n
Most important features<\/p>\n
Let’s continue by looking at the most important features according to two different tests. Than we will use the top ones to train and test our first model.<\/p>\n
In\u00a0[18]:<\/strong><\/p>\n
\n
#First we need to split the dataset in the y-column (the target) and the components (X), the independent columns.<\/span>\r\n#This is needed as we need to use the X columns to predict the y in the model.<\/span>\r\ny =<\/span> data['Customer Lifetime Value'<\/span>] #the column we want to predict<\/span>\r\nX =<\/span> data.<\/span>drop(labels =<\/span> ['Customer Lifetime Value'<\/span>], axis =<\/span> 1<\/span>) #independent columns<\/span>\r\n<\/pre>\n<\/div>\n

\nIn\u00a0[19]:<\/strong><\/p>\n
\n
from<\/span> sklearn.feature_selection<\/span> import<\/span> SelectKBest\r\nfrom<\/span> sklearn.feature_selection<\/span> import<\/span> chi2\r\n#apply SelectKBest class to extract top 10 best features<\/span>\r\nbestfeatures =<\/span> SelectKBest(score_func=<\/span>chi2, k=<\/span>'all'<\/span>)\r\nfit =<\/span> bestfeatures.<\/span>fit(X,y)\r\ndfscores =<\/span> pd.<\/span>DataFrame(fit.<\/span>scores_)\r\ndfcolumns =<\/span> pd.<\/span>DataFrame(X.<\/span>columns)\r\n#concat two dataframes for better visualization<\/span>\r\nfeatureScores =<\/span> pd.<\/span>concat([dfcolumns,dfscores],axis=<\/span>1<\/span>)\r\nfeatureScores.<\/span>columns =<\/span> ['Name of the column'<\/span>,'Score'<\/span>] #naming the dataframe columns<\/span>\r\nprint<\/span>(featureScores.<\/span>nlargest(10<\/span>,'Score'<\/span>)) #print 10 best features<\/span>\r\n<\/pre>\n<\/div>\n
\"Text<\/p>\n
In\u00a0[20]:<\/strong><\/p>\n
\n
#get correlations of each features in dataset<\/span>\r\ncorrmat =<\/span> data.<\/span>corr()\r\ntop_corr_features =<\/span> corrmat.<\/span>index\r\nplt.<\/span>figure(figsize=<\/span>(20<\/span>,10<\/span>))\r\n#plot heat map<\/span>\r\ng=<\/span>sns.<\/span>heatmap(data[top_corr_features].<\/span>corr(),annot=<\/span>True<\/span>,cmap=<\/span>\"RdYlGn\"<\/span>)\r\n<\/pre>\n<\/div>\n
\"Timeline<\/p>\n
What pop’s out when looking at the correlations for the CLV is the column ‘Monthly Premium Auto’ and the ‘Total Claim Amount’ These might be the best features to use.<\/p>\n
Seems that the feature selection models differ a bit in which feature is the most important. For the first test I will keep:<\/p>\n