Thursday 23 January 2025
A team of researchers has developed a new approach to predicting demand for products in situations where supply is limited. The method, called Diffused Censored Gaussian Process (DCGP), uses machine learning algorithms to take into account both the availability of products and customer behavior.
The DCGP model works by simulating how customers behave when their preferred product is out of stock. It assumes that customers will look for alternative products in nearby locations, and it uses this information to estimate the demand for each product.
To test the effectiveness of the DCGP model, the researchers used data from three different sources: supermarket sales, bike-sharing systems, and an artificial dataset generated to mimic real-world scenarios. The results showed that the DCGP model outperformed traditional methods in all cases, even when the train set size was reduced.
The researchers also experimented with varying the parameters of the DCGP model, including the diffusion lengthscale (ℓdiff) and sink node probability (πdiff). They found that the performance of the model was robust to misspecification of these parameters, but that learning πdiff from the data improved results.
In addition, the researchers tested the DCGP model on a version of the bike-sharing dataset with only two hubs. The results were consistent with those obtained for the full dataset, further demonstrating the versatility and effectiveness of the approach.
Overall, the DCGP model offers a promising solution to the problem of demand prediction in situations where supply is limited. Its ability to simulate customer behavior and take into account both product availability and demand makes it a valuable tool for businesses seeking to optimize their inventory management strategies.
The researchers’ findings have significant implications for industries that rely on predicting demand, such as retail and logistics. By using the DCGP model, these companies can better anticipate fluctuations in demand and make more informed decisions about inventory levels and supply chain management.
In the future, the researchers plan to continue refining the DCGP model and exploring its applications in other domains. They hope that their work will contribute to a better understanding of customer behavior and improve decision-making processes in various industries.
Cite this article: “New Approach to Demand Prediction in Limited Supply Situations”, The Science Archive, 2025.
Demand Prediction, Supply Chain Management, Machine Learning Algorithms, Diffused Censored Gaussian Process, Customer Behavior, Inventory Management, Bike-Sharing Systems, Supermarket Sales, Artificial Dataset, Robustness Analysis.
Reference: Filipe Rodrigues, “Diffusion-aware Censored Gaussian Processes for Demand Modelling” (2025).
