Reviewer has chosen to be Anonymous

Overall Impression: Good

Content:
Technical Quality of the paper: Good
Originality of the paper: Yes, but limited
Adequacy of the bibliography: No

Presentation:
Adequacy of the abstract: Yes
Introduction: background and motivation: Limited
Organization of the paper: Needs improvement
Level of English: Satisfactory
Overall presentation: Good

Detailed Comments:

The paper proposes a hybrid AI approach to a domain-specific prediction problem on time series. An ensemble of neural models takes center role, whereas rules are employed for error detection and correction. Datasets are explicitly referenced, ablation is reported.

The beginning of the paper seems to suggest that the selection of metals (Cobalt etc.) is somehow relevant to the problem statement. I am not sure this is the case, or how is this prediction problem influenced by the nature of the goods to which it was applied. Is there anything that makes this framework specific, is it generalizable as a time series analysis approach? Is MPSC really a problem class by itself, if yes what domain-specific aspects or temporal features make it distinguishable? Or is it just an application case determined by the availability of the datasets? Elaborating on the nature of the problem, e.g. of spikes to be detected could strengthen the argument in this respect.

The main criticism however is that the special issue to which this was submitted emphasizes a role of conceptual modeling in neurosymbolic AI. While the work is clearly relevant for the journal as a hybridization of complementing AI approaches, and there is some domain-specificity manifesting in the problem statement, the conceptual modeling aspect is rather absent/implicit and should be brought into the discourse explicitly - either as an ingredient of the technical proposition itself (a metamodel for rule modeling could be an interesting approach), or as a frame for the problem analysis or solution orchestration, or at least an ontological meta-view on the framework as a whole (i.e. where the neural networks are first class modeling constructs in relation/dependencies with the other involved ingredients). In the absence of any effort in this direction I don't see how the paper can be relevant for the special issue.

The bibliography is too shallow and consequently the Related Works section too brief, considering the diversity of existing preoccupations with rule mining - see https://arxiv.org/abs/2408.05773. The paper should position itself against this landscape.

Finally, because this appears to be an extended version of a conference paper, the conference version should be explicitly referenced, and a clear explanation of the nature and extent of the applied extension must be provided in the beginning of the paper.
See the journal policy on this: https://neurosymbolic-ai-journal.com/content/author-guidelines