System and method for predicting customer lifetime value using two-stage machine learning

A system, wherein the system comprises a computing device, the computing device comprises a processor and a storage device storing computer executable code, and the computer executable code, when executed at the processor, is configured to:

• provide a classifier of a computing device, wherein the classifier is trained using customer feature data of a first plurality of customers during a first period of time as input and whether there are spendings of the first plurality of customers during a second period of time as classifier labels;
• provide a regressor of the computing device, wherein the regressor is trained using customer feature data of a second plurality of customers during the first period of time as input and amount of spendings of the second plurality of customers during a second period of time as regressor labels, wherein the second plurality of customers is a subset of the first plurality of customers, and the amount of spending of each of the second plurality of customers during the second period of time is positive;
• perform the classifier using customer feature data of a third plurality of customers during a third period of time to obtain predicted classifier labels;
• obtain a fourth plurality of customers, wherein the fourth plurality of customers is a subset of the third plurality of customers, and the predicted classifier labels of the fourth plurality of customers being positive;
• and
• perform the regressor using the customer feature data of the fourth plurality of customers during the third period of time to obtain predicted regressor labels, the predicted regressor labels being customer lifetime values (CLVs) of the fourth plurality of customers,
• wherein training the classifier by using the customer feature data of the first plurality of customers during the first period of time as the input and whether there are spendings of the first plurality of customers during the second period of time as the classifier labels comprises:
• retrieving customer features of the first plurality of customers from customer data and transaction data during the first period of time, wherein the customer features of the first plurality of customers comprise orders, browses, clicks and add-to-cart features of the first plurality of customers, and demographic information of the first plurality of customers;
• aggregating the orders, browses, clicks, and add-to-cart features of the first plurality of customers to obtain aggregated features, wherein the aggregated features comprise aggregated spendings of orders;
• normalizing the aggregated spendings of orders to obtain normalized features;
• encoding the demographic information of the first plurality of customers to obtain encoded features;
• combining the aggregated features, the normalized features, and the encoded features to obtain the customer feature data of the first plurality of customers during the first period of time;
• generating 1 or 0 as the classifier labels based on whether there are the spendings of the first plurality of customers during the second period of time;
• and
• training the classifier by using the customer feature data of the first plurality of customers during the first period of time and the classifier labels during the second period of time.

A non-transitory computer readable medium storing computer executable code, wherein the computer executable code, when executed at a processor of a computing device, is configured to:

• provide a classifier of a computing device, wherein the classifier is trained using customer feature data of a first plurality of customers during a first period of time as input and whether there are spendings of the first plurality of customers during a second period of time as classifier labels;
• provide a regressor of the computing device, wherein the regressor is trained using customer feature data of a second plurality of customers during the first period of time as input and amount of spendings of the second plurality of customers during a second period of time as regressor labels, wherein the second plurality of customers is a subset of the first plurality of customers, and the amount of spending of each of the second plurality of customers during the second period of time is positive;
• perform the classifier using customer feature data of a third plurality of customers during a third period of time to obtain predicted classifier labels;
• obtain a fourth plurality of customers, wherein the fourth plurality of customers is a subset of the third plurality of customers, and the predicted classifier labels of the fourth plurality of customers being positive;
• and
• perform the regressor using the customer feature data of the fourth plurality of customers during the third period of time to obtain predicted regressor labels, the predicted regressor labels being customer lifetime values (CLVs) of the fourth plurality of customers,
• wherein training the classifier by using the customer feature data of the first plurality of customers during the first period of time as the input and whether there are spendings of the first plurality of customers during the second period of time as the classifier labels comprises:
• retrieving customer features of the first plurality of customers from customer data and transaction data during the first period of time, wherein the customer features of the first plurality of customers comprise orders, browses, clicks and add-to-cart features of the first plurality of customers, and demographic information of the first plurality of customers;
• aggregating the orders, browses, clicks, and add-to-cart features of the first plurality of customers to obtain aggregated features, wherein the aggregated features comprise aggregated spendings of orders;
• normalizing the aggregated spendings of orders to obtain normalized features;
• encoding the demographic information of the first plurality of customers to obtain encoded features;
• combining the aggregated features, the normalized features, and the encoded features to obtain the customer feature data of the first plurality of customers during the first period of time;
• generating 1 or 0 as the classifier labels based on whether there are the spendings of the first plurality of customers during the second period of time;
• and
• training the classifier by using the customer feature data of the first plurality of customers during the first period of time and the classifier labels during the second period of time.