Blockchain-based federated learning with smart contract for fruit waste prediction

Abstract

Fruit losses in the supply chain owing to improper handling and a lack of proper control are common in the fruit logistics industry. As losses are caused by the inefficiency of the export method, selecting the appropriate export method is a possible solution. Several organizations employ only a single strategy, which is mainly based on a first-in-first-out approach. Such a policy is easy to manage but inefficient. Given that the batch of fruits may become overripe during transportation, frontline operators do not have the authority or immediate support to change the fruit dispatching strategy. This study aims to develop a dynamic strategy simulator to determine the sequence of delivery based on forecasting information projected from probabilistic data to reduce the amount of fruit loss. The proposed method using asynchronous federated learning (FL) is based on blockchain technology and a serially interacting smart contract. In this method, each party in the chain updates its model parameters and uses a voting system to reach a consensus. This study employs blockchain technology with smart contracts to serially enable asynchronous FL, with each party in the chain updating its parameter model. A smart contract combines a global model with a voting system to reach a common consensus. Its artificial intelligence (AI) and Internet of Things engine further strengthen the support for implementing the Long Short-Term Memory (LSTM) forecasting model. Based on AI technology, a system was constructed using FL in a decentralized governance AI policy on a blockchain network platform. With mangoes being selected as the category of fruit in the study, the system improves the cost-effectiveness of the mango supply chain. In the proposed approach, the simulation results show fewer mangoes lost (from 6.91% to 0.035%) and operational costs reduced.