How to address food insecurity using analytics modeling
Food redistribution using Analytics Modeling
Target
To redistribute edible, consumable food that has been discarded by retailers to people in need, reducing the hunger problem while lowering the negative environmental effects of food wastage.
Where are we now?
Food wastage is one of major environmental and social issues we face in North America. USA is a global leader in food wastage summing up to ~$161 billion dollars of food wasted every year which is estimated to be almost 131 billion pounds of food per year¹. Approximately 9% of US population (37 million people) is food insecure today. On one hand $1600 dollars of food is wasted per person per year and on the other hand, average food needs are not being met by ~37 million Americans².
Food is wasted at every stage of the supply chain, that is, raw materials at farms, at the industries, at wholesalers and retailers, at restaurants and at consumer level. Of all the food that is uneaten only less than 3% is redistributed and everything else results in waste. A part of it is used for Composting but mostly all of it goes to landfills, while also resulting in environmental loss as its not just the end-product being wasted, but also the resources required for preparation (water, land, trees, etc.)
Unfortunately, the retailers (supermarkets) alone contribute 43 billion pounds of food wastage every year.
If you were to see this problem through the lens of operations research, wouldn’t you want to use right away some of the advanced analytics models to reduce the operating costs and improve the efficiency that it becomes more profitable project? My approach follows this path to ideate a solution that can become a possible reality — here is how I would do it.
Where are we going?
Scope -
To solve this problem, let us first narrow our focus on specific food items discarded by retailers like dented cans, blemished packages, blemished fruits and vegetables, wrong sized products, overstocking, unpurchased holiday foods. After the project success, we can scale it up to include more consumer food products and also increase our reach to more food insecure communities.
Solution in a nutshell
1. Gather the demographic data, implement a clustering model to find the clusters of rich and poor communities.
2. Contract with the retail food stores in the rich communities to take their good-faith food donations of their discards.
3. With the list of participating stores, set up redistribution centers nearby. Start collecting food and take it to the redistribution centers.
4. Improve efficiency and reduce operational costs of the food processing using queuing model at the redistribution centers.
5. Reduce the risk of distributing unconsumable food by using a neural network model to identify and separate expired food items from the good ones.
6. Finally, achieve efficiency in operations by using an optimization model to redistribute food to food insecure communities.
How are we going to get there?
Efficiency mindset -
Let’s say we were able to gather enough demographic data for a specific region on map, either through the Census Bureau or through surveying. Do some market research to find the rich and poor communities. If you want to be certain then use the k-means clustering model to find the clusters and their centroids of rich and poor communities. With this knowledge, narrow the focus on retail stores which are more likely to generate more waste (most likely in richer communities).
With the list of supermarkets and grocery stores willing to participate in this great initiative by agreeing to do good-faith food donations through their discarded foods, we can establish redistribution centers closer to them. We have already managed to reduce the travel time and duration for the discarded food collection at redistribution centers.
AI for a good cause –
We don’t want to give bad or expired food to the hungry. For this, as the items arrive at the redistribution centers, they will be scanned by their bar codes to get the basic details. A neural network model will be built to identify the expiry dates on the products when they are scanned and organize them in bins of expiry duration bands (range). This filtering will make sure only good food gets out for redistribution.
Operational efficiency –
These redistribution centers need to have just enough workers that it can be operated with minimal operational costs. A queuing model can be a savior here. Built to project the right number of workers needed based on the arrival rate of the food and service rate of the workers. This will help to run the process smoothly so that the food doesn’t go bad waiting to be processed.
The final and the most important step — redistribution. Once the system is operational and food is ready to be distributed, an optimization model will be implemented to minimize the overall cost of delivering the food to multiple centroids of poor areas by taking the food with maximum duration to expire to the farthest locations. This will also account for bringing back the untaken food to the redistribution center. Ideally, a food truck with frozen food will optimally distribute the food on specific schedules to locations where people can come to pick food.
More details on Modeling
Model 1: K-means Clustering
Given: Annual Income, Location Latitude, Location Longitude
Use: K-means clustering
To: Find the centroids of clusters of working class, rich and poor people
Data Collection:
Data can be used from government provided sources like the Census Bureau. Additionally, promotional or paid surveys can be run to gather the necessary information for process initiation.
Model 2: Convolutional Neural Network
Given: Product Image, Expiry Date, Expiry Type
Use: Convolutional Neural Network (CNN) model
To: Classify the Expiry Type and determine the Expiry Date
Data Collection:
Using UPC barcodes each food product’s details can be retrieved. For food products that do not have a UPC code, details should be entered manually. This will be relatively small number of items like fruits and vegetables. Food will be scanned to take a photo of the expiry label.
Model 3: Queuing Model
Given: Arrival rate (λ), Service rate (µ)
Use: Queuing model (Call center)
To: Predict the number of workers needed for scanning and processing food items in queue
Data Collection:
Arrival time of the food items on the conveyer belts should be recorded either by a machine or manually. This will give us the output which will be fit against a probability distribution. Service time for each food product to be scanned and photo to be taken must be recorded by the system.
Model 4: Optimization
Given: Food Expiry Duration, Expiry Date, Expiry Type, Food Type, Product Name, Product Size, Product Price, Available Workers For Distribution, Distribution Distances
Use: Optimization model
To: Minimize the overall cost of redistribution
Data Collection:
As we have most of the data from previous models we will append data about the different distribution locations using the centroids from K-means for poor areas. The distances will be calculated from Redistribution center to each location. The associated standard travel cost will be calculated per trip.
References
1. The Estimated Amount, Value, and Calories of Postharvest Food Losses at the Retail and Consumer Levels in the United States
https://www.ers.usda.gov/webdocs/publications/43833/43680_eib121.pdf?v=0
2. Feeding America: Food Insecurity
https://hungerandhealth.feedingamerica.org/understand-food-insecurity/
3. Model Good Samaritan Food Donation Act
https://www.govinfo.gov/content/pkg/PLAW-104publ210/pdf/PLAW-104publ210.pdf
4. The Census Bureau Income data
https://www.census.gov/topics/income-poverty/income.html
Bazil Ahmed is a Data & Analytics consultant at Slalom and a Data Science student at Georgia Institute of Technology. I believe in creative thinking and sharing my research/ideas with the community. This is my very first step towards this journey of knowledge sharing. Here’s my LinkedIn if you would like to connect with me. If you would like to learn more about Data Science at Slalom please feel free to reach out to me @bazil.ahmed@slalom.com
