Sistema de previsão do preço do milho em Moçambique baseado em machine learning

Abstract

The present work aims to develop and implement a web-based system for forecasting maize prices in the Mozambican market, using Machine Learning techniques and artificial intelligence, introducing a modern solution as a differentiating approach compared to traditional methods. With the ongoing growth of web platforms that provide public data, institutions such as NGOs have been making data on Mozambique available, including historical records of the agricultural market. These increasingly abundant and accessible datasets contain key variables necessary to address the growing need for accurate forecasts of maize price fluctuations in the country. Thus, an in-depth study of Machine Learning strategies based on statistical models becomes essential. Identifying the most appropriate model with optimised parameters and reliable results is a significant challenge to ensure a comprehensive solution to the research problem. In a set of conducted experiments, three machine learning algorithms were compared, namely Multiple Linear Regression, LSTM and Random Forest, applied to historical data on Mozambique’s maize market. The LSTM model was selected as the most effective and was adopted as the core forecasting engine of the proposed system. The final system integrates a backend developed in Django, which provides a functional web platform for visualising predictions. Additionally, it incorporates an AI-based chatbot designed to interact with users, clarify questions, explain forecast results and facilitate intuitive and conversational access to information. The combination of robust models and high-quality data provides a pathway for improving the accuracy of market dynamics forecasting, promoting economic stability and supporting better management of agricultural resources.