<p>The surge in demand for cloud computing services leads to increased energy consumption and massive carbon emissions in geographically distributed datacenters. Therefore, cloud service providers (CSPs) seek eco-friendly solutions to reduce datacentres’ energy consumption and carbon emissions. One such solution is powering the datacenters using renewable/green energy to minimize the usage of non-renewable/brown energy. Renewable energy generation heavily depends on the datacenters’ geographical location and time, and its proper management can reduce carbon footprints to a greater extent. In this context, recent literature forecasts the renewable energy and the power consumption concerning each datacenter’s location and time and accordingly manages the load using the autoregressive integrated moving average (ARIMA) machine learning (ML) model. However, this model requires too much data preprocessing and tuning using trial and error to determine the optimal parameter values for predicting central processing unit (CPU) utilization. This paper minimizes the data preprocessing and tuning by introducing another ML model, the backpropagation neural network (BPNN), in a renewable energy-based virtual machine (VM) placement algorithm. The algorithm considers energy and carbon footprints to place VMs effectively by maximizing the usage of green energy and minimizing the carbon footprint. The proposed algorithm is extensively simulated on 16 servers distributed across four datacenters in Denver, Portland, San Francisco and Albuquerque locations to accommodate a set of VMs. The simulation results of the proposed algorithm are compared with the energy and carbon footprint aware with the predictive (EFP) algorithm (uses ARIMA model) in terms of energy consumption in watts, carbon footprints in kilogram (kg) and non-renewable energy usage in watts to show the effectiveness.</p>