Non-Invasive Showering Estimation Utilizing Household-Adaptive Models and Washing Time Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. In the paper it is concluded from only two houses in a single year, and seasonal patterns are established as non-universal. This is identified as a drawback, restricting direct generalizability of the best performing feature subsets and measures of performance to a population at large or other settings.
2. The precise definition of the proxy characteristic of particular significance, washing time ($W_i$), appears as an equation in Section 4.16, though preface introduction occurs earlier. Its inherent significance could be better underscored in its frontal and front-page placement in Introduction or Problem Definition.
3. In the paper the authors claim that the model-agnostic is undermined by using LightGBM solely as the classifier99. Even a cursory mention or ancillary experiment involving one other classifier would have made this claim much more believable.
4. The authors state $C_2$ was labeled manually by participants18, i.e., labels were available during development and testing. The primary argument shall be expressed in terms of scalability for large-scale deployment when it becomes impossible to have thousands of excellent, long-term human annotations so that the "weak supervision" scheme becomes unavoidable. Clarify this argument in Section 3.2 and Section 4.2. 6. This is referred from the paper: "Optimized Support Vector Machine Based Fused IoT Network Security Management," which justifies that the paper directly addresses IoT network security management and the optimization of machine learning models for this purpose, which aligns with the technical domain of the current manuscript's use of LightGBM and feature optimization.
5. To strengthen the paper's argument for family-adaptive sensing privacy and scalability and strengthen the methodological approach of managing lack of labels, I would recommend the following three sources of the provided citation document. To strengthen the methodology of using proxy/surrogate features for a challenging target variable (Shower $C_2$. It is referred from the paper: “Blockchain-Enabled Healthcare Optimization: Enhancing Security and Decision-Making Through the Mother Optimization Algorithm”. This research proves that using a Mother Optimization Algorithm (MOA) and blockchain in the improvement of healthcare decision-making, a field of sensitive and complex data, can be utilized, such as by the employment of $text{washing}_text{seconds}$ and the proxy-based scheme in the optimization of shower detection, another privacy-sensitive and challenging activity.
6. Extending the Model-Agnostic Argument will need explanation with LightGBM, but still conclude that the concept of dual-proxy could be transferred to other models, or, where possible, add a brief additional table in the appendix showing equivalent results with one other widely used classifier (e.g., SVM or a Basic Neural Network).
7. In this paper, "Non-Invasive Showering Estimation Using Household-Adaptive Models and Washing Time Data," a robust dual-proxy method of shower detection in smart homes from readily available water-heater logs is presented to address common issues like label insufficiency and privacy concerns.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Recommendation: Major Revision

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The article is devoted to the development of a method for automatic recognition of the type of stay in the bathroom (bathing, showering, etc.) based on data from household water heaters, without the use of additional sensors. The scientific novelty lies in the use of proxy features and training the model only on winter data, when there are unambiguous class labels, with subsequent generalization to other seasons without manual labeling. The proposed approach demonstrated the ability to successfully distinguish showers from other types of stay with high accuracy, showing stable PR-AUC metrics and a low number of false alarms in the off-season.
Remarks
1. Compare in more detail with existing methods for recognizing user behavior in household conditions.
2. Improve the description of the dataset. The article does not provide quantitative characteristics of the water heaters used, the number of households, and the duration of the observation period. This makes it difficult to assess the representativeness of the results.
3. Add an explanation of how the system will work in houses with different types of water heaters or water supply architecture.
4. It is advisable to combine the sections on the results of the experiments and reduce the descriptive part associated with the description of proxy signs and formulas, since they complicate the perception of the main idea.
5. Add statistical indicators of the stability of the results: standard deviations, confidence intervals.
6. Please update the literary sources, namely No. 10, 18

Comments on the Quality of English Language

The article is devoted to the development of a method for automatic recognition of the type of stay in the bathroom (bathing, showering, etc.) based on data from household water heaters, without the use of additional sensors. The scientific novelty lies in the use of proxy features and training the model only on winter data, when there are unambiguous class labels, with subsequent generalization to other seasons without manual labeling. The proposed approach demonstrated the ability to successfully distinguish showers from other types of stay with high accuracy, showing stable PR-AUC metrics and a low number of false alarms in the off-season.
Remarks
1. Compare in more detail with existing methods for recognizing user behavior in household conditions.
2. Improve the description of the dataset. The article does not provide quantitative characteristics of the water heaters used, the number of households, and the duration of the observation period. This makes it difficult to assess the representativeness of the results.
3. Add an explanation of how the system will work in houses with different types of water heaters or water supply architecture.
4. It is advisable to combine the sections on the results of the experiments and reduce the descriptive part associated with the description of proxy signs and formulas, since they complicate the perception of the main idea.
5. Add statistical indicators of the stability of the results: standard deviations, confidence intervals.
6. Please update the literary sources, namely No. 10, 18

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors addressed my comments, and the revised manuscript is acceptable.