Is it possible to predict only by categorical features in ML of which there are more than 100?

Question

Machine learning

Let's say predicting the cost of a house for 200 features, where each individual house will have only about 10 features (10 non-zero values ​​in the vector). Are such predictions practical?

Answer 1

Any method is not a panacea. The answer to your question is that it is possible to predict. But whether there is a dependence of your target feature on your (even 100) independent features is a completely different question.
Failure in forecasting can be caused by both an incorrectly chosen method and the lack of correlations between input and output features.
And, by the way, the selection of the most adequate features is one of the most important tasks in ML.

Answer 2

Do you mean that one particular house may have 200 features, but in general there are only 10 features, and the other 190 features are empty?
Then the question arises, is there any consistency from house to house in which signs are empty and which are not? Or will one house have non-empty signs conditionally x1 - x10, and another x150-160?
The very essence of cost forecasting lies in the fact that some key features are selected that affect the cost. There must be some feature x5 (or rather, several such features) that all or almost all houses will have, so that the variability of this feature can be associated with price volatility.
In any case, at first glance, the data looks very sparse, which clearly does not have a positive effect on the construction of any dependencies.
It would also be nice to know about the number of records.
Try to make a couple of basic models, suddenly something will come of it.