Replenishment optimization using ML
This was my approach to Hack2Vax Hackathon.
https://www.infostretch.com/hack2vax-hackathon/
Problem Statement:
The primary objective of this Hackathon is to promote team building, creative problem solving, and the use of digital engineering services to address the challenges faced in the covid-19 vaccine distribution
supply chain.
Abstract:
The impact of the Covid-19 on supply chain management has also been seen in vaccination distribution as well. However, by considering over a short period of time quality check and production planning have improved. Here we are trying to solve the overstocking and/or understocking problems by replenishment optimization using machine learning and data science techniques. With this approach, we can also minimize transportation costs and reduce storage space (rental cost). Thus, leading to a cost-effective vaccine distribution to everyone and on time
About:
In any pandemic, vaccination plays a vital role. Since vaccination is perishable, its entire lifecycle should be managed carefully. When it comes to storage, it shouldn’t be under/overstocked. Due to various factors, the data available is very minimal, based on this data it is really a challenging task to design a solution around predictions.
Hence assuming that pre-requisites are met, like production of the vaccine, quality check and storage quality, etc. We are focusing on the under/overstock challenge which is termed as Replenishment Optimization in SCM terminology.
The process of optimization starts with the prediction of vaccine quantity required and total vaccinated numbers
Data:
Reference to the data collected from:https://www.kaggle.com/c/demand-forecasting-kernels-only/data we have modified as per the current Covid-19 scenario for the year 2021 (vaccine is available only this time) only.
Regarding data features:
‘date’, — Date of vaccine dd/mm/yyyy
‘hospital’,-Distibution Center (10)
‘item’,-Type of vaccine (Covi-shield,covaxin, etc)(3)
‘qty’-Total vaccine used on that day (in Lakhs)
Records =54780
Sample data looks like this:
Machine Learning approach:
In this problem initially, we will create some time-series features to see how the trends are impacted by day of the week, hour, time of year, etc. And analyze. Further, with vaccine quantity forecasting and prediction using the XGBoost model.
The predictive model helps in inventory optimization (Minimizing the number of deliveries from the Warehouse)gives a strong strategy for replenishment optimization
GitHub Repo: https://github.com/MadhuriKonnur/replenishment_optimization
<Note: This is modified data of retail sales. Due to scarcity of actual data>
Vaccine Distribution Model Prediction visualization:
Future scope:
- Observing the date patterns, it is better to consider the holiday calendar
- Demand Planning & Delivery Schedule Scope example 4 days or replenishment per week: Monday, Wednesday, Friday, Sunday 24 hours lead-time between order creation and delivery from the warehouse Store opening from 10:00 am to 08:00 pm Delivery from the warehouse at 09:00 pm
- Further creating Data: Create Sales DataFrame, Inventory Record, Capacity by Item, and Replenishment Data
-Experimenting with other algorithms for interesting analysis and results
Simple deployment:
References:
https://www.latentview.com/blog/how-machine-learning-can-improve-covid19-vaccine/
