*2.1. Low-Frequency Datasets*

The following low-frequency datasets were analyzed and compared: Smart [18], HES [19], Tracebase [20], DataPort [21], AMPds [22], iAWE [23], GREEND [24], REFIT [25], and RAE [26].

The relevance of these datasets is due to the characteristics of the installed measuring devices. However, many of the possible strategies for feature extraction, that can be used for NILM classification, are restricted due to low sampling frequency.

A comparison between some characteristics of the different types of low-frequency NILM datasets can be seen in Table 2. Where *fs* represents the sampling frequency; DCD stands for Data Collection Duration; NoC corresponds to Number of Appliance Classes; NoA represents the Number of Appliances; and Res., Lab., Com. and Ind. are short forms for: Residential, Laboratory, Commercial and Industrial installations, respectively. Since the sampling occurs at very low rates (once a minute to once a second) the recordings can take place for very long times (weeks to years).


**Table 2.** Comparison between low-frequency Non-Intrusive Load Monitoring (NILM) datasets.
