Hi,

no one can tell you what is the exact number of observation or datapoint in the training dataset to build a robust model, it depends on the dimention of your input and the machine learning model used. however you need a enough of observations to get good generalization and acceptable acurracy

I am not sure, but if there is a lack of data, maybe techniques as bootstart can be adequated.

the number of samples you need depends on the complexity of the analysis performed, the chosen algorithm and the method of sample selection for validation of the model (leave one out, cross-validation, holdout, bootstrap ...), the minimum acceptable performance, of the number of explanatory variables .... That is to say, I believe that it is not possible to define a priori. From the practical point of view the number of samples collected depends, among other things, on how much time and how much financial resources are available. A good literature review on the subject may give some referential.

learn with small dataset

https://www.quora.com/What-are-some-machine-learning-algorithms-that-can-learn-well-even-with-a-small-data-set

How much training data required for machine learning

https://www.quora.com/What-are-some-machine-learning-algorithms-that-can-learn-well-even-with-a-small-data-set