Electricity consumed by residential consumers counts for a significant part of global electricity consumption and utility companies can collect high-resolution load data thanks to the widely deployed advanced metering infrastructure. There has been a growing research interest toward appliance load disaggregation via nonintrusive load monitoring. As the electricity consumption of appliances is directly associated with the activities of consumers, this paper proposes a new and more effective approach, i.e., activity disaggregation. We present the concept of activity disaggregation and discuss its advantage over traditional appliance load disaggregation. We develop a framework by leverage machine learning for activity detection based on residential load data and features. We show through numerical case studies to demonstrate the effectiveness of the activity detection method and analyze consumer behaviors by time-dependent activity modeling. Last but not least, we discuss some potential use cases that can benefit from activity disaggregation and some future research directions.