Genetic Algorithm and Evolutionary Architecture for Optimizing Stochastic Processes – A Supply Chain Use Case

Abstract

Product placement is one of the most complicated process in supply chain and logistic management for eCommerce industries. It requires prioritizing which product to be placed and in which channel to optimize product delivery time and maximize profit. There can be no perfect decision in this matter but can be chosen from a wide array of possible solutions which in turn represents only a subset of a big and complex solution space. This important decision has impact downstream on product delivery, product return, customer satisfaction and profitability of the eCommerce platform. This can also have impact on churn and other business drivers. We have taken up this problem and tried to apply Genetic Algorithm (GA) to come up with a best fit solution. In this heuristic study we will also propose a reference public cloud-based deployment strategy to deploy the solution proposed in a public cloud environment. Its no new news that machine learning applications are now widespread and often an integral part of the primary software that’s supporting one or more business processes. With the advent of new machine learning and deep learning paradigms, developing and applying models to solve business problems have become comparatively easier. This fueled with cloud-based platform support for big data has created an ecosystem of increasingly complex predictive solutions. Such solutions when deployed at a large scale have intrinsic problems associated with dynamic software systems, viz., complication in architecture at scale and all-around observability. Deployment of a system that has predictive algorithms at the end of numerous data pipelines coupled with end point management for consumption by downstream applications and visualization platforms suffers from problems such as scalability and security. My paper goal is to explore these problems in a growing ecosystem of predictive analytics project and propose microservices architecture for predictive models as using containerization and container orchestration.