There are a lot of machine learning models, such as SVM, LR, etc.

First, you must be applying 90% at least from the models, then you can decide to enhance or develop or hybrid .to find the best model for your data prediction.

recommend :

Flood Prediction Using Machine Learning Models: Literature Review.

"https://doi.org/10.3390/w10111536"

A general framework for forecast analysis and evaluation can be found here:

Chapter Forecast Evaluation Techniques for I4.0 Systems

The idea is to try alternative approaches and then to select the most suitable approach. It's important to define the criteria for choosing the best option, and the above chapter provides a detailed guide to this task.

For the flood prediction methods, firstly, of course you'll need to define the quantity you need to predict.

Then the framework from the chapter above describes the following workflow:

1) prepare the data using the TSTS;

2) implement alternative models and store forecasts using the FTS format;

3) use queries to obtain data in the AFTS format, use the variant of the PRD-diagram proposed on p. 89, use the AvgReMAE metric to compare alternative models, pick the best model.

Note that sophisticated models can be costly and benefits from improved modelling should outweigh the costs of data preparation and collection for the advanced modelling approaches.

I think you start by recent literature review discussing the major challenges, threats, and opportunities of flood prediction. From the literature, you will find the key insights as the starting point. Extra points, Mustafa Mhamed Abdalaa recommends Machine learning review with sufficient algorithms can be enhanced, optimized using meta-heuristic and evolutionary algorithms.

I echo the suggestions / answers listed above. Start by looking at literature reviews will give an idea on how to go about the entire data mining process. But before that, to tackle the requirement of your master degree, you may want to first check if you are able to obtain the appropriate flood dataset for your project to safe time.

My suggestion departs from the norm. Do you have some causal understanding/model in mind to explain when flooding occurs, when it doesn't, and when it stops occurring? To me it is the addition of water beyond what is absorbed or removed. I'm no flood scientist, but I can confidently say floods are CAUSED by water flux in > water flux out--conservation of mass, that is water mass. Anyone who fills a bath tub knowingly/unknowingly is applying this principle.

A non-causal model is not very useful in my mind since it's probably unintelligently trying to get to the same root of the problem without consciously seeking a causal link. A model that makes good predictions may be causal, but not necessarily so (think Ptolemy), and may have embarrassing/costly moments of inexplicable model failures; sometimes this is better than nothing, but avoid if you can.

As far as data collection, the flux problem (input and output of water) depends on the region your evaluating. Where does water enter and leave the area? What are the limits of absorption/drainage? Can they be measured? If so this ought to be your data collection strategy. Check out weather underground for a source (rain) of water influx: https://www.wunderground.com/history.

For a master's degree it may be best to keep it simple, select an area where rain is the only influx of water, and drainage is the only outflux--this would be deserts, ya they flood, Tucson AZ does once a year pretty much. Looking at past floods (in the news archives), and associated rain rate (inches/unit time) ought to get you an upper limit. Also looking at when floods did NOT occur and its associated rain rate out to get you a lower limit. If this proves to be too simple, then gradually add complexity, maybe look at regions with streams and therefore rain/snow-melt upstream matters, or look at regions with absorption properties and therefore recent history of rain matters.