CASAS Smart Home dataset - scripted complex activities with injected cognitive-related errors

This dataset represents ambient data collected in the CASAS smart apartment testbed at Washington State University. In this study, 328 performed scripted activities while ambient sensors continuously collected readings. Some of the scripted activities were completed as specified. In other cases, participants injected an error into the activity. Note: Other CASAS smart home and smartwatch datasets are also available, look for more at https://zenodo.org/communities/casas. The activities are: Cook. Take medicine. Select an interview outfit. Make a phone call. Water plants. Sweep the floor. Play cards. Watch a DVD on the television. The types of errors that were injected are commonly encountered by individuals with cognitive impairment. The types of errors are: Omission: Skip a step of the task. Substitution: Use an alternate object not commonly associated with the task, or use an appropriate object in an unusual (ineffective) way. Irrelevant: Add a step that is unrelated to the activity and unnecessary for completing the activity. Inefficient: Add a step that slows down or compromises the efficiency of the activity. Participants performed the activity without error, then performed the activity again with an error. They were given freedom to select their own action that fit the specified error type. The RawData folder contains the sensor readings for each participant (one file per participant). The floorplan of the apartment with sensor locations is provided in the file Chinook.jpg. Sensors in the apartment are categorized (and named) as: M01 - M51: PIR motion detectors (ON when detected motion starts and OFF when it stops) I01 - I10: item use sensors (PRESENT or ABSENT indicating item is on sensor or not) D01 - D019: door sensor on cabinets and doors (OPEN or CLOSE) P001 and P002: current eletricity consumption T001 - T006: ambient temperature sensors BATP and BATV: sensor battery levels The file ErrorTypes_AllActivities.csv contains extracted features describing the activity errors, with the corresponding activity and error type. The ErrorAnnotation.csv file provides additional information about each error and how it is detected. In the Data_IndividualActivities directory, these data are further divided into activity, participant, training (no error) and testing. There are five types of extracted features that define the errors: sensorID: the sensor associated with the most recently-reported sensor reading at the time of the error support: ratio between number of participants that triggered the most recent sensor and the total number of participants eventNumProb: probability of the number of sensor readings from the activity beginning until the most recent sensor reading eventNum: the number of sensor readings from the activity beginning until the most recent sensor reading eventNumProb: probability of the elapsed time from the activity beginning until the most recent sensor reading eventTime: elapsed time from the activity beginning until the most recent sensor reading eventPause: the time elapsed between the previous and current sensor reading sensorPause: the time elapsed between the previous and current reading from the most recent sensor sensor counts (multiple): number of readings so far from each sensor probability sensor counts (multiple): Poisson probability of the current distribution of readings across sensors