Using LSTM Neural Networks as Resource Utilization Predictors: The Case of Training Deep Learning Models on the Edge