Abstract

When measuring the supply–demand gap, this article takes into account a variety of factors, including product categories, classifications, and spatiotemporal characteristics. For the purpose of closing the supply–demand gap, we developed and implemented a full system that makes use of an extendable deep neural network architecture. The framework is capable of analysing a variety of custom input items and automatically detecting supply and demand trends based on transaction data from previously completed transactions. A generic training model is developed to estimate future demand based on a collection of customizable attributes, which is then tested. In order to consolidate input data, embedding layers are used to map high-dimensional features onto a smaller subspace, resulting in a more compact representation. The model's training architecture is composed of completely connected layers with activation functions that are then coupled together. Custom data attributes can be concatenated from many levels of the deep learning neural network in order to create a more complex model.