Recently my manager asked to consider machine learning in our company. The specific problem he had in mind could be described as follows.
There are customers with datasets with employee information - gender, nationality, age, marital status, last salary increase time, current salary, CV information (can deduce "how many times the employee has changed their job in last 10 years" etc.) and also we could include general job market data (typical salary for this position in other companies etc.). Based on all of these features and their historically observed results, we would like to train ML models that can predict specific simple yes/no answers about employees, such as "are they considering leaving the company?", "are they undervalued?" and generate a report on employees who have high probability scores for predicted "yes" answers.
I'm an experienced .net C# developer and also have general architecture experience with Azure (VMs, app services, functions, DevOps) but I have no serious experience with machine learning yet. Some years ago I was playing around with Nvidia's StyleGAN and neural networks based speech synthesis, but I was treating the "AI stuff" as a black box, tweaking only the control UIs and data input utilities. However, yesterday I watched a few MLNET tutorials and it all seemed to make sense and made me thinking that even a "mere mortal" .net developers might be able to create something usable, especially if we get some help from a data scientist sometime later.
Would MLNET Model Builder be enough to help with this specific scenario? Would binary classification model be a good candidate? Or maybe Azure ML Studio might offer a better starting point?
Are there any code examples that show how to properly featurize such kind of input data and that might work well for our case, at least to have a usable proof-of-concept to demonstrate? Or am I oversimplifying the task and it would need much more data analysis and algorithm selection than automated ML.NET and Azure builders can offer?