Prediction customization

Hello everybody,

here I want to describe in more details what I mean when I say prediction customization.

As name says by itself, I want to make something that will help you to make predictions. 

Predictions for what? 

Acumatica has plenty of entities: invoices, sales orders, employees, taxes, acounts payable, acounts receivable, fixed assets, etc. I want to make customization, that will use some machine learning techniques in order to say something like: in the next month based on history of your your sales, you'll have 2000 sales orders. Or based on your current history, you'll pay 300 000 taxes in the next two months. Or based on your history of hiring, in next three years you'll need to search for a new office. Examples are numerous.

Why?

To make good poanning. Main reason why somebody may need such a customization involves planning. I give you one example. I worked for company which as usually is in top 30 of fortune 500 list. I used to be a subcontractor of this company. Do you know, what kind of planning that company had? I worked for that company at 2016, but that company had a plan of activities up to 2021 which means they had plan for 5 years ahead. Another example which you can consider. Imagine, that to you comes your sales manager, and says to you: we will need a lot of such and such items on the next month. You ask him, how much stands for a lot? And then sales will start guess work in order to say how much. 

How it works?

There are plenty of mathematical models which can help you to tell what will be in the future. For mine customization I want to use three main models:

  1. Random forest
  2. xgboost
  3. Neural networks

Of course there are plenty of other ways of modelling the future events, but I would like to start from the first two, and later add Neural Networks. 

Why not start with Neural networks right away?

Neural networks is quite costly solution from the viewpoint of hardware greediness. Especially if to think from standpoint of training neural network. In order to train a good model it can be needed to have good pc with powerful video card. And as for today powerful video cards are expensive because quite often they are used for bitcoins generation.

Summary

If you wish to add to your Acumatica machine learning capabilities, please let me know. I'll be happy to add AI to your Acumatica and make your toool smarter.

No Comments

Add a Comment