Hello, everyone,
I want to use markov model to predict continuous time-series data, and I read some references, It seems that the model is more suitable for discrete data , so I divided the continuous time-series data into discrete time-series data first, namely, divided the continuous time-series data into different states, and then use markov model to predict it,
but, I will generate different states too, the predicted data is still discrete data, and how should I transform the state data into continuous time-series data, the length is as long as the origional continuous time-series data?
And if there is a real data example and codes using markov model running in matlab or R? I searched some codes running in matlab and R, but I failed.
Your valuable suggestions are most welcome .
Dear Saiping Jiang
I am not an expert in statistics, but I have come across a PDF file in the Google.
In that PDF document, there are some concise review of the Markov model concepts. There, with regard to a Markov model, you may read:
- Discrete outputs are modeled using pmfs (probability mass functions).
- Continuous outputs are modeled using pdfs (probability density functions).
I have attached the PDF along with a screen-shot for your perusal.
Hope it would be helpful.
Yes, you can. What you are looking for is the Chapman-Kolmolgorov transition equation. You can then derive the Fokker-Planck differential equation which allows you to solve for a time dependent probability distribution. You can find an outline of the derivation and use of the Fokker-Planck equation in some specific examples here:
Technical Report Markov Chains for Fun and Profit: From Gambler's Ruin to Pha...
To use the first-order Markov model predicting continuous time-series data, you have to use the Chapman-Kolmolgorov transition equation. After that, you have to derive the Fokker-Planck differential equation which allows you to solve for a time dependent probability distribution.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series