**5. Conclusions**

This comparative study is an analysis of the various ML methods used in land use planning along with their advantages and disadvantages. The study also has brought out the applicability of ML algorithms to specific task of land use planning. Some of the articles deal with the testing of the performance of specific ML algorithms and while others have dealt with specific requirements of land use planning. In particular, the aspects of urban land use change, relevant to understand the spatio-temporal patterns of urbanization are prominent in the literature. To a lesser extent, the socio-political reasons and causes of these patterns are part of these studies, although it can be imagined that ML algorithms could also address part of these via other types of data repositories. The review of various publications on ML algorithms and their applications in land use planning demonstrates that most of the ML algorithms operate at the pixel level and have limited practicality when working with VHR imagery, in contrast to object-based image analysis. In general, non-parametric algorithms are computationally more expensive compared to linear models. The efforts are more towards minimizing the complexity and lowering the computational cost. The comparison of the various algorithms indicates further that random forest is relatively more robust for classification. The choice of the ML methods relies largely on the type of datasets and understanding of the researcher. It has been seen that hybrid approaches like pixel based and object-based approaches are gaining popularity to ge<sup>t</sup> the optimum/desired results for modeling urban growth, land use change, settlement patterns and classification.

The benefit of this review is that it enables the boundary work between practitioners and future researchers in either land use planning or in the application of technical spatial and information scientific tools. The overview tables present the links.

Further research should go in the direction of utilizing empirical data sets and applying the algorithms to analyze their performance and suitability for studying aspects of land use planning. Experiments on deep learning, SVM, random forest and GANS for land use classification should check the overall classification accuracy of each of the algorithms. In addition, experiments with spectral, texture, structural and contextual image features should be undertaken to improve the accuracy of classification. The performance of the hybrid approach should be tested to ge<sup>t</sup> best results in classification, modeling of urban land use growth, change and transition.

**Author Contributions:** Conceptualization, V.C. and W.T.d.V.; methodology, V.C. and W.T.d.V.; investigation, V.C. and W.T.d.V.; writing—original draft preparation, V.C. and W.T.d.V.; writing—review and editing, V.C. and W.T.d.V.; supervision, W.T.d.V. Both authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by TUM open access publishing fund.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
