I’m not really sure which machine learning approach is best for my problem at hand. I work in an engineering company that designs and builds different kinds of ships. In my particular job, I collect the individual weight of items on these vessels. The weight and there location is important because it is used to ensure the vessel in question can float in a balanced manner.
I have a large corpus of historical data on hand that lists the items on the vessel, there attributes, the weight for these items and where the weight came from (documentation), or the source.
So, for example, let’s say I have the following information:
ITEM | ATTRIBUTES | WEIGHT |WEIGHT SOURCE Valve |Size: 1 inch |Type: Ball Valve |2 lbs. |Database 1 Elbow |Size: 2 inch |Type: Reducing |1 lb. |Database 2
I have to comb through many systems on these vessels and find the proper documentation or engineering drawings that lists the weight for the item in question. It usually starts by investigating the item and its attributes and then looking in a number of databases for the weight documentation. This takes a long time, as there is no organization or criteria as to what database has what. You just have to start randomly searching them and hope you find what you need.
Well I now have a large corpus of real world data that lists thousands of items, there attributes, there weight and most importantly the source of the documentation (Database 1, 2, 3 etc.). I’m wondering if there is any correlation between an item, its attributes and its weight location (database). This is where machine learning comes in. What I’d like to do is use machine learning to help find the weight location more quickly. Ideally it would be nice if it could analyze a batch of information and then provide recommendations on which databases to search.
My first thoughts are that this is a classification problem, and maybe a CNN would be helpful here. If that is the case, I have over 100 categories in my dataset. I actually went ahead and programmed a simple feed forward neural network using the following resources: https://www.analyticsvidhya.com/blog/2017/05/neural-network-from-scratch-in-python-and-r/ I attempted to use this network to solve the above problem, but so far I have had no success. I’m in over my head here.
I don’t expect it to be correct 100% of the time. Even if it had an 80% success rate that would be awesome. So my question is this;
What kind of neural network do I need to accomplish this?