I'm working on some ideas for a project for work. I'm pretty new at machine learning/predictive analytics(I'm 50% way through a udacity class).
I'm going to use a made up example, to protect my company, but hopefully this gets my question across.
For my job, we take in customer requests, and the fulfil those orders. Our work consists of many steps to complete the order, including an installation of equipment at the customers office. Let's say that we install security systems.
The customer submits a request, which we then provide them a quote and they decide if they want the system. That is the first part of the process. If they want the system, we then start the provisioning process. At this time, we provide them an estimate of how long everything will take. This is a fairly simple deterministic system based on past data.
We want to move to a more complex system to provide them a better estimate. And one that updates as milestones are completed.
In our process, I'll say we have 10 milestones that are done to complete the order. Some can be done in parallel, but not all. We have information about the type of system ordered, where they are in the world, and several other categorical descriptive features. We have transnational data on the orders to say when each step was completed and other events about each step in the process. We generally have been collecting data 3 times a day from our systems to capture changes as they happen. We do have holes in the data, where something failed to run every once in awhile.
Again, our goal is to predict for new orders, how long it will take to complete the order.
What model types should I research? I feel like this is a time series problem, but without enough experience, wanted to get some thoughts.