The Data Mining Forum 
IMPORTANT: This is the old Data Mining forum.
I keep it online so that you can read the old messages.
Please post your new messages in the new forum: https://forum2.philippe-fournier-viger.com/index.php
Sediment Analysis using AI techniques
Posted by:
Koen Wildemeersch
Date: March 05, 2014 12:01AM
Hi all,
I have this question about a the following research topic we are tackling in Flanders hydraulics research institute (Flemish Government, Belgium): we have gathered a lot of sediment concentration measurements along the Scheldt River (Belgium) as well as the dredging dumps that took place in the river Scheldt.
We’ve collected this data for four years 2008-2012. The sediment measurements were made at a very high resolution: every 10 minutes. For the dredging dumps we have only the day the dumping took place as well as the quantity. This data is available at different locations in the Scheldt and thus would allow us to find a tempo-spatial relation between sediment and dump.
My question goes as follows: How would you tackle the problem? We would start off with a correlation analysis and would than consider performing a regression analysis (ANN, SVR) to predict sediment concentration as function of tide, season, … and dump quantity. The idea we have is to use an train an autoregressive model. Would you agree?
One issue is that a lot of dumps take place during that 4 year. So it’s a very complex relation that is observed.
I hope I managed to explain the scope of the work in a way that is easy to understand for you. I would have preferred to add a figure, but as far as I know this is not possible.
Anyway do not hesitate to contact me for further info. All input is very much appreciated.
Kind regards,
Koen
I have this question about a the following research topic we are tackling in Flanders hydraulics research institute (Flemish Government, Belgium): we have gathered a lot of sediment concentration measurements along the Scheldt River (Belgium) as well as the dredging dumps that took place in the river Scheldt.
We’ve collected this data for four years 2008-2012. The sediment measurements were made at a very high resolution: every 10 minutes. For the dredging dumps we have only the day the dumping took place as well as the quantity. This data is available at different locations in the Scheldt and thus would allow us to find a tempo-spatial relation between sediment and dump.
My question goes as follows: How would you tackle the problem? We would start off with a correlation analysis and would than consider performing a regression analysis (ANN, SVR) to predict sediment concentration as function of tide, season, … and dump quantity. The idea we have is to use an train an autoregressive model. Would you agree?
One issue is that a lot of dumps take place during that 4 year. So it’s a very complex relation that is observed.
I hope I managed to explain the scope of the work in a way that is easy to understand for you. I would have preferred to add a figure, but as far as I know this is not possible.
Anyway do not hesitate to contact me for further info. All input is very much appreciated.
Kind regards,
Koen
Re: Sediment Analysis using AI techniques
Posted by:
Koen Wildemeersch
Date: March 13, 2014 05:49AM
Nobody? All information is welcome really.
Kind regards, Koen
Kind regards, Koen
Re: Sediment Analysis using AI techniques
Posted by:
webmasterphilfv
Date: March 13, 2014 08:04AM
Hi,
I think that it is an interesting problem and that you have interesting data, which is great. When I had read your post for the first time, I did not understand the meaning of dredging but I have searched a little bit on Google and now i understand and I see that you want to analyse the relationship between dumping and the sediment in the river that would be the result of these dumps.
I think that it may be interesting to try models like ANN and SVM. It could give you a rough idea about whether you can get some good results or not. Then, perhaps that you could try some other methods that are specifically designed to handle time and compare the results with ANN and SVM.
If you have 4 years of data, a simple way to test your models would be to split your data in half. Then you would train your model with the first two years and check if the results allows to predict the next two years, for example. Some more complicated way to test your model could be to use k-fold cross validation.
This is just my opinion. Actually, I'm not an expert on problem with time series, so maybe someone else could also provide additional information.
Best,
Philippe
Edited 1 time(s). Last edit at 03/13/2014 08:05AM by webmasterphilfv.
I think that it is an interesting problem and that you have interesting data, which is great. When I had read your post for the first time, I did not understand the meaning of dredging but I have searched a little bit on Google and now i understand and I see that you want to analyse the relationship between dumping and the sediment in the river that would be the result of these dumps.
I think that it may be interesting to try models like ANN and SVM. It could give you a rough idea about whether you can get some good results or not. Then, perhaps that you could try some other methods that are specifically designed to handle time and compare the results with ANN and SVM.
If you have 4 years of data, a simple way to test your models would be to split your data in half. Then you would train your model with the first two years and check if the results allows to predict the next two years, for example. Some more complicated way to test your model could be to use k-fold cross validation.
This is just my opinion. Actually, I'm not an expert on problem with time series, so maybe someone else could also provide additional information.
Best,
Philippe
Edited 1 time(s). Last edit at 03/13/2014 08:05AM by webmasterphilfv.