Diffusion modeling starting with time series data

tgb417 Dataiku DSS Core Designer, Dataiku DSS & SQL, Dataiku DSS ML Practitioner, Dataiku DSS Core Concepts, Neuron 2020, Neuron, Registered, Dataiku Frontrunner Awards 2021 Finalist, Neuron 2021, Neuron 2022, Frontrunner 2022 Finalist, Frontrunner 2022 Winner, Dataiku Frontrunner Awards 2021 Participant, Frontrunner 2022 Participant, Neuron 2023 Posts: 1,595 Neuron

Hi all,

Is there anyone out there that is doing diffusion modeling (uptake, innovation) modeling in Dataiku starting with time series data of similar products.

Use case I working with perishable products that have an expectation date. Once the date is past that particular product or service is no longer available. Others like it may or may not be made available.

We do have a bunch of time series data for older products which are similar in qualities. these items include:

  • Quantities of product of each type made available in ranging from:
    • fixed groups of a few hundred to a few thousand individual instances of the product to more product that the market can use.
  • Other features about the product provided by Subject Matter Experts that in their experience are considered important like:
    • brand name
    • recency that a similar product might have been made available.

I’m interested in hints from anyone about possible approaches

I’ve done some looking at:

  • Bass Diffusion Modeling, that seem to produce the right kind of general model, but may tend to under estimate values. Estimating the parameters p, q, m are proving a challenge for me.
  • There are variants to Bass Diffusion Modeling like:
    • Gompertz function
  • Arima forecasting. Which does not seem to decompose the non repetitive nature of the data.

Thanks for any ideas.

Operating system used: Mac OS Ventura 13.3

Setup Info
      Help me…