Exploring Low-Cost Machine Learning Methods for Rainfall Forecasting to Empower Less Developed South African Communities

Heavy rainfall is a weather phenomenon where an area receives high measures of precipitation (in the form of rain) in a relatively short period of time. It can cause a series of adverse risks, such as death, economic loss and infrastructure damage. While weather forecasting strives to reduce these risks by predicting these events before they occur, the forecasts provided by the government describe the risk to the full area that the heavy rainfall will affect. However, rainfall disproportionately affects less developed areas. With the increase in computation power in recent history, machine learning has become a viable method to improve the existing weather forecasting techniques due to its ability to handle vast amounts of complex data. However, the models that the government use are highly complex and expensive. As such, this study aims to use a quantitative machine learning approach to investigate the efficacy of utilising low-cost machine learning to allow communities to make accurate rainfall predictions in their local area, allowing them to evaluate the risk that the rainfall poses in their local context and enabling informed risk prevention decisions.