A Project Example: Direct Mail Planning

Direct Mail Planning System

A Project Example

Using Machine Learning methods to increase direct mail efficiency.

Industry: Marketing

Project description:

Our customer sends out direct mail asking for donations. This project is aimed at optimizing mailing efforts, taking into account that all letters are sent via a post service and each letter has a non-zero cost.

Tasks implemented:

Use of Cost-sensitive reinforcement learning method.
This approach concentrates on building an optimal sequence of decisions with the goal of maximizing total benefits accrued over a period of time instead of immediate benefits.
In the scope of this project the popular Markov Decision Process model with value function approximation was adopted.
Development of a scalable Reinforcement Learning framework capable of running on Hadoop / Apache Spark clusters using Apache Spark’s MLlib machine learning library
Creating an app to demonstrate the functionality of the new environment. The demonstration is based on data related to direct mail statistics from public sources.

Algorithms and mathematical methods:

Reinforcement learning, Sarsa, Q-learning
Logistic and Linear Regression
Decision Trees
Cost-Proportionate subsampling

Development environment:

Cluster framework: Hadoop, Apache Spark
Programming language: Scala
Math libraries: MLlib, Breeze
Database: HBase
Operating Systems: Linux, Windows

Project size:

6 thousand lines of Scala code
2 months, 3 developers