The aim of this study was to analyze the impact of current temperatures in major US cities on the accuracy of forecasts for natural gas futures prices using various machine learning algorithms. The study involved comparing the results of six algorithms (decision trees, random forests, support vector machines, gradient boosting, k-NN, and neural networks) in two versions: without temperature data (Study A) and with this data (Study B). The results indicate that the inclusion of temperature data significantly improves prediction accuracy in all tested algorithms. The best results were achieved by gradient boosting, showing the highest accuracy (0.581) after adding weather data. The analysis of variable importance revealed that minimum and maximum temperatures, as well as the “day_of_year” variable, were key predictors of gas prices. The conclusions of the study suggest that incorporating temperatures into predictive models may be beneficial for “end-of-day trading” investment strategies, leading to more accurate investment decisions. The study's limitations include the limited scope of weather data and the need for further testing under various market conditions. These results can help investors and market analysts develop more effective investment strategies.