Climate Change and Global Warming: A Machine Learning Study Forecasting Global Land Temperature to 2050

This report covers a machine learning study of global climate change and warming, built on two publicly available datasets: the Berkeley Earth Surface Temperature (BEST) dataset and the Our World in Data (OWID) CO2 Emissions dataset. The idea was simple to state but harder to do properly. We wanted a model trained on 265 years of temperature and emissions data that could actually survive scrutiny, not just post a good score. Along the way we also wanted to know which parts of the world are heating up fastest, and to put together a defensible forecast for global land temperature in 2050. We followed a fairly cautious process. Four raw CSV files were cleaned one at a time instead of forced through a single pipeline. Features were engineered from 1,992 monthly records. Five candidate models were compared against each other before we settled on one. The first attempt, a Random Forest Regressor, returned an R2 of 0.9902 on the surface. That number looked great until we checked what was actually driving it. Month alone accounted for over 95 percent of the model's feature importance, meaning it had basically learned the calendar instead of the climate. It also could not extrapolate past 2015. We dropped it. The model we kept was a Ridge Regression trained on six engineered features. It reached a test R2 of 0.9927, a mean absolute error of 0.2805 degrees Celsius, and a cross-validation standard deviation of 0.0005. Running it forward month by month to 2050 gave an absolute land temperature of 10.18 degrees Celsius. That works out to roughly 2.28 degrees of warming since 1850, which sits comfortably inside the range the IPCC has already projected. On the geographic side, Mongolia came out on top at +1.836 degrees Celsius, with Kazakhstan and Kyrgyzstan close behind at +1.464 and +1.423 degrees. Ulan Ude and Chita led the city-level rankings. Taken together, these results support the idea that CO2-linked features are not just statistically convenient. They line up with the physical mechanism that is actually driving the warming.