Machine learning is revolutionizing various fields, but its implementation in real-time soft environments often faces challenges due to limited computational and storage resources. In this work, we have successfully developed a highly accurate Random Forest regression model to predict the working ambient temperature for an embedded Radio Access Network system, particularly within the Baseband application domain. Our model achieves minimal prediction error and maintains a variance well aligned with the onboard sensors' measurement accuracy. Remarkably, the outcomes of our research respect the stringent real-time processing and storage constraints, making it a significant advancement in real-time machine learning applications.