The Case for AI in Food Waste

A team of 5 AI apprentices from AIAP Batch 11 took part in the recently concluded Smart City Ideation Challenge and emerged as the champion for the category “Sustainability & Carbon Neutrality”. The Challenge aimed to seek insights from the global AI community on “What are the AI problems worth solving towards Smart City Sustainability?”

Their winning proposal which was titled “The Case for AI in Food Waste” considered the use of Federated Learning for food demand prediction.

We caught up with them in this interview to find out more…

Team Name: The Ape-prentice
Team Member(s): (L-R) Teo Wei Yi, Darren Lee, Angus Saw, Feng Wei Han, Justin Lim



  1. Congratulations on winning the AI competition! Can you tell us a bit about your team and your roles within the team?
    Our team was comprised of individuals from the AI Apprenticeship Programme (AIAP). Instead of assigning specific roles, we divided ourselves into smaller groups to focus on different report sections, as there was limited time. Darren was chosen to represent our team to present our solution during the Final Pitch.
  2. What motivated your team to participate in this AI competition? Were there any specific goals or challenges that you wanted to tackle?
    Our team’s motivation for participating in the AI competition stemmed from our shared passion for AI and its real-world applications. The challenge provided an opportunity for us to showcase the knowledge and skills we acquired during the apprentice program, as well as address various goals and challenges.
  3. Why the name Ape-prentice? Anything to do with apes? 🙂
    The name “Ape-prentice” is a play on words. “Apprentice” refers to our involvement in the apprentice programme, while the “Ape” component is inspired by our proposed federated learning solution and the movie “The Planet of the Apes.” The famous line, “Apes together strong,” accurately represents our objective—enabling small business owners to collaborate and leverage the power of AI collectively.


  1. Can you provide an overview of the solution that your team developed for the competition? What were the key techniques or algorithms that you employed?
    Unfortunately, the competition was just an ideation so there wasn’t any implementation or development. Our team proposed the use of Federated Learning for food demand prediction. Federated learning is a decentralised approach to training a model, which has been widely employed in situations where data privacy is crucial, such as in healthcare. Our innovation comes from the use of federated learning for food demand prediction – allowing business owners, even competitors, to harness the power of AI together. We also proposed unique evaluation metrics that can help the organizer to evaluate the model performance.
  2. How did the team prepare for the competition? Did your team conduct any research or experiments beforehand to fine-tune your approach?
    There were 2 different problem statements to tackle: sustainability and fake news. After researching current solutions, we decided to focus on sustainability and specifically target food waste. We promptly applied the knowledge we gained from an AIAP senior sharing session on federated learning to our specific use case. We were initially unfamiliar with the technique, so to better understand federated learning, we conducted a similar experiment using a toy dataset we found online to test the concept’s feasibility. We spent the remaining days learning as much as we could about federated learning and incorporating our newfound knowledge into the report.
  3. Were there any limitations or trade-offs in your solution? If so, how did your team address them?
    Our solution certainly had its limitations and trade-offs. We aimed to develop a general model capable of predicting food demand for a wide range of items, from basic bread to dishes in high-end restaurants. As one can imagine, the demand for these two extremes can differ significantly. Given the heterogeneity of the target variables, achieving this goal immediately wasn’t feasible. To address this challenge, we proposed a phased approach. In the first phase, the focus would be on predicting the number of transactions, which food stall owner could then use as a proxy to estimate the amount of food needed for preparation. Once successful, we would implement incremental improvements to eventually reach item-level predictions. Additionally, we proposed incorporating store-level data, such as ratings, location, and type of food, to help the model learn the characteristics of various food stores or restaurants. This information would aid the model in making more accurate predictions.
  4. What are the potential real-world applications or implications of your winning AI model beyond the competition? How do you plan to further develop or deploy it in the future?
    The proposed AI model can be used to forecast demand in food establishments. By adopting this solution, companies can benefit by reducing costs through better inventory management and manpower allocation. It also can reduce the greenhouse gas emitted from wasted food, where one-third of the food produced globally is wasted. We could use real-world data from food establishments to train a time series model and perform feature engineering and hyperparameter tuning on the model as well. During deployment, a pipeline should be set up where data from the POS is updated regularly (e.g., daily/weekly) to train the local and global models which are hosted on cloud services.



  1. What was the most exciting or memorable moment for your team during the competition? Did you encounter any unexpected surprises or breakthroughs?
    The most memorable moment was when our team stayed till 10pm one day to look through our proposal in detail and to finalise it. It became clear that we would need to work harder than we initially anticipated. However, seeing our team working towards a common goal and eventually winning the competition made all the hard work put in truly worthwhile.
  2. How did your team handle collaboration and communication throughout the competition? Did you encounter any difficulties or learn any lessons about teamwork during the process?
    We distributed the workload in a way that leveraged the strength of each team member and demonstrated a high level of dedication throughout the competition, regularly checking in with one another to provide updates on our progress and address any issues that arose. We were able to overcome challenges in our assigned tasks through effective communication and a willingness to work together to find solutions. These factors helped to keep us all on the same page and to work towards the same goal as a team.
  3. How did your team approach the problem differently from other participants? Were there any unique insights or strategies that you employed to gain an edge in the competition?
    We chose to apply AI to tackle the food waste problem in a preventive setting instead of the conventional approach of recycling. By proposing a solution which can allow food establishments to make more accurate demand forecasting, it can stop potential waste from being produced in the first place. Furthermore, the federated learning approach is a rather novel but apt application to the problem.
  4. What were the most significant challenges that your team faced during the competition, and how did you overcome them?
    One of the most significant challenges was adapting the federated learning solution to the context of the competition requirement, which asked for a proposal that could be implemented as an AI challenge similar to a Kaggle challenge. We had to conduct research into various metrics that are specific to federated learning, which has important considerations for the privacy preserving capabilities of the model. In the end, we came up with a combined weighted metric that takes into account various performance considerations such as predictive capabilities and privacy preserving capabilities of the model.



  1. What did you think were the key success factors pertaining to this win?
    We think that a few factors contributed to our success in the competition. Firstly, in terms of the problem impact, we chose to focus on reducing the waste generated by local food establishments by proposing a solution which would allow them to perform better demand forecasting. This solution, if implemented at scale, has the potential to benefit both big and small food establishments, offering a complementary approach to the typical recycling strategy. Our proposal tackles wastage before it is generated, which makes it an innovative and impactful solution. Secondly, we think that our proposed solution, which makes use of federated learning for demand forecasting, is a novel application of federated learning, which is itself a new but increasingly relevant field in machine learning. By making use of federated learning, individual food establishments could come together and build a better model collectively without privacy worries, which can potentially be a real concern given the sensitive nature of sales data. Finally, the style of presentation was clear and understandable. An anecdotal approach was also used to allow audiences to relate to the business use case. Overall, we think that these few factors came together nicely from ideation to presentation and produced a rather coherent project.
  2. How did being in AIAP help with the win?
    Being in AIAP allowed us to build a solid foundation in various areas of machine learning, including exposing us to federated learning, which was used in the proposed solution of this project. It allowed us to gain the foundation to understand machine learning research papers and ground our proposal in solid research. Furthermore, all the group members got to know each other through the AIAP, it is fair to say that the project would not have been possible without AIAP!



  1. What advice do you have for aspiring data scientists or AI practitioners who want to participate in AI competitions and improve their skills?
    Staying up to date with the latest developments in the AI/ML field is essential to understanding the various approaches that can be employed to tackle the problem. Then focus on a specific area and conduct extensive research, depending on the competition, one may even want to do some form of quick EDA and modeling to test the feasibility. Finally, being able to communicate your solution to technical and non-technical audiences alike is also important.
  1. Lastly, what does winning this AI competition mean to your team? How do you feel about your accomplishment and what are your plans or aspirations in the field of AI?
    Winning this AI competition is certainly a confidence booster, especially for most of us who are still rather new to the AI field. As this competition is mainly an ideation competition, we hope to take part in more projects and challenges in the future where we can really implement AI solutions that can have a real-world impact. This competition and the AIAP are important stepping stones and opportunities for us to build on.

Link to presentation:

Darren (last one on right) seen here with representatives from the winning teams together with the Technical Review Committee Members