While gentrification can regenerate neighbourhoods, it can also exacerbate economic and social disparities. Gentrification forecasting is therefore essential to establish a comprehensive picture of neighbourhood deprivation and development. The employment of machine learning models provides an opportunity to conduct gentrification forecasting beyond traditionally applied regression models. This study considers three types of ML models (random forest, gradient boosting machine, and extreme gradient boosting), analyses their predictive performance against each other, and compares them to a traditional regression approach. The study aims to forecast gentrification in London in 2025. The findings show that the tuned extreme gradient boosting achieves the highest predictive accuracy. They do not confirm that all machine learning models outperform traditional regressions. The most accurate model’s forecast yields that 66.53% of neighbourhoods experience no significant change, while 20.33% are forecasted to gentrify and 13.14% are projected to undergo a relative increase in deprivation. The features found to highly influence all models were unemployment, tenure, education, and income levels.