Edge-cloud systems are heterogeneous computational infrastructures designed to manage workloads across distributed environments. Accurate performance predictions of these systems are essential for optimizing resource allocation, reducing latency, and improving overall efficiency. Traditional performance analysis tools, which do not leverage machine learning (ML), often introduce significant errors. This study proposes the use of ML models to predict key performance metrics in heterogeneous edge-cloud systems. Five different ML models are trained and used to estimate the system’s performance. The models include Random Forest (RF), Long Short-Term Memory (LSTM), Deep Neural Network (DNN), Recurrent Neural network (RNN), and a hybrid model combining RNN and DNN. The input datasets for training are sourced from the Wichita State University Computer Architecture and Parallel Programming Laboratory (CAPPLab). The performance of these models is evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The predicted communication latency and power consumption values are then compared against previously generated VisualSim results to assess the accuracy and effectiveness of the ML-based approach. The simulation results indicate that the RF model delivers superior performance, as reflected in its lower testing MAE and RMSE, demonstrating its effectiveness in capturing system dynamics. The deep learning models—LSTM, DNN, and RNN—exhibit varying levels of accuracy, with some models outperforming others in specific scenarios. The hybrid RNN-DNN model strikes a balance between computational complexity and predictive accuracy. The ML-predicted values follow a similar trend to the VisualSim results.