AI Series – Part 5: Modeling Intro

At the Modeling Stage, you should have accomplished the following actions:

  • Acquired clear, authentic, organized data which will serve as training data for the base of your project. (3VofData: volume, variety, velocity)

  • Decided on leveraging OOB capabilities to limit customizations.

  • Interpreted the patterns of the data via a user-friendly visual format: flows, charts, maps, etc. (*visualization tools: PPT, Tableau, Google or Lucid charts, dashboards, etc)

  • Identified the suitable lifecycles (use cases) and outputs (metrics), gap analysis, stages, etc, so that you can measure the stages of success of the project. (Review ‘Predictive Modeling’ exercise for foundations)

Modeling is best determined by “output of a problem that will drive business value in a reasonable amount of time”. The project should be a visual, value based and timely to gain momentum and trust from your teams.

Now you can begin to select your model(s) based on the above work and choose the most efficient one to begin with. Your primary model will become the base framework of your AI project. Review the Predictive Model (article) to build your foundation, team and narrow the gap for success before jumping directly into AI.

The next step is to have your team select the appropriate algorithms, tuning parameters and a set of processes for continuously evaluating the models to find the best fit.

It is also important to validate the models to ensure that they are accurate and reliable and this includes validating the data along the project journey. How do you know you have chosen the right models? data? and definitions of truth (or non-biased data)? Ever asked 5 different people at an organization what defines success or what should be measured? Hint: same thing. More on this later.

I work with field assets so I have to work backwards from the Asset input (IoT or Field Service Tech via Dynamic Checklists) to the output (Analytics) back to input (Predictive Models) and eventually, where we are headed now – AI. My goal is for the company to build a better relationship with their field assets which means more quality data. Pretty straight forward. My question is simple as well. What do I need to do to build a better communication with the asset?

Note: For any AI project to be successful, the training data should be relevant to the problem statement and authentic so that your outcome is real. Ex. Don’t use data off a random site as accuracy cannot be proven.

2 Options: Rule based or Learning Based

Rule Based requires a developer to push the data along with the rules to the machine which will predict answers based on training. The issue is that it is not flexible, and is not antifragile but fragile as it is constant (vs dynamic) in design. It gives what it gets, no changes are applied.

Learning Based (Deep Learning and Machine Learning) means that the machine model to put it simply is dynamic and learns by itself and so becomes an antifragile system. As data is pushed to the system model, it makes new associations from the data and adapts.

The model you choose depends on the problem or challenge required.

This was high level. Deep diving into modeling next.

Previous
Previous

AI Series – Part 4: Predictive Modeling

Next
Next

AI Series – Part 6: The Trifecta