Figure 5.
The process is titled Factors Influencing the Energy Consumption. The Pre-Training Phase includes energy required for business understanding, data understanding, data preparation, and other activities. The Training Phase includes training data, such as its amount and structure; the machine learning algorithm used, including hyperparameter tuning, pretrained models, and G e n A I; I T infrastructure; and the A I or machine learning service model. Re-training and fine-tuning can repeat within this phase. The Usage Phase includes the amount of use, such as single or batch usage, and the I T infrastructure used, including its related lifetime. Backward loops can return from the Training Phase to the Pre-Training Phase and from the Usage Phase to the Training Phase when necessary. These phases lead to energy usage and carbon dioxide emissions as the environmental impact.Final framework of environmental impact of machine learning services based on findings in literature and recommender system observations as well as expert evaluations