Sensor-Based Human Activity Recognition Using Adaptive Class Hierarchy

Round 1

Reviewer 1 Report

This paper investigates adaptive class hierarchy in training of CNN model for sensor-based activity recognition. The authors proposed a method to automatically create a class hierarchy from the training data and train the B-CNN using the created class hierarchy. Their method performs classification considering the hierarchical relationship between classes without prior knowledge. The experimental results on three datasets indicated the effectiveness of the B-CNN model. Overall, the paper is clearly motivated and is technically sound. Minor concerns are:

1. Can the the proposed solution be adapted to video-based action recognition? If yes, what are the differences compared to those solutions for video-based action recognition?

2. Can the proposed model provide an interpretable prediction? please clarify. (see for example:  "Visually interpretable representation learning for depression recognition from facial images," IEEE TAC, 2020.)

3. While the paper focuses on the sensor-based action recognition, it is necessary to cite and discuss some recent works on video-based solutions with class hierarchy.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The comments are as follows,

1. In 4.4 Evaluating Model, the authors use three datasets. Why the partition of HASC is different from WISDM and UniMib SHAR?
2. 2 shows the basic model structure you use. But why the authors use 5 max pooling layers and convolution layers with 3 classifier? Is it the best setting for all datasets? The authors may add more experiments to show how the parameters of model affect the results, for example change the number of convolution layers.
3. In related work, there many deep learning methods to solve this problem. The authors might add more baselines to compare with your model.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I have no further comments.