Text this: RTLA-HAR: a model proposal based on reinforcement and transfer learning for the adaptation of learning in human activity recognition