Table 4.
Analysis of various windowing methods for HAR.
| Methods | Idea | Advantage | Disadvantage |
|---|---|---|---|
| ABW | The data stream of events is divided into windows at activity change detection points | (i) Suitable for labeling data. (ii) If valid points are detected, the accuracy of detection increases. |
(i) Failure in activity recognition correctly (ii) Suitable for online recognition (iii) Complexity of calculations in finding practical separation points (iv) Inaccuracy in the boundaries of activities |
|
| |||
| TBW | Event data streams are divided into windows with fixed time intervals | (i) The simplicity of implementation. | (i) Choosing the right window length (ii) Extremely influential window length in decision making |
|
| |||
| SDW | The data are split into windows with the same number of sensor events, and the results of window time lengths vary from window to window | (i) This approach offers computational benefits over ABW. (ii) No need for sensor events to classify past sensor events. |
(i) There may be a significant time interval between an event and the previous event (ii) Performance is low in the face of two or more residents in a smart home (iii) Giving equal importance to all data (iv) The possibility of having a window containing sensor events for a long time (v) The possibility of having a window containing sensor events related to the transfer between two activities |
|
| |||
| SDEW | The mutual information of the two sensors described earlier depends on the order in which a pair of sensors occurs in the entire data stream | (i) Uses multiple sensors to increase detection accuracy. | (i) The possibility of losing some dependence between the sensors (ii) In parallel activities and sensor events, one activity can be described for other information (iii) Dependence between sensors |
|
| |||
| L-SSW | In a window specified by the Ai event sensor, a sensor can be activated several times | (i) Sometimes, the latest sensor status, according to ei, can be more descriptive than the frequency at which it occurs in a window. (ii) Simplicity of calculations. |
(i) There may be a significant time gap between an event and previous events (ii) Challenges more than one person living in a smart home |