Hi, Thanks a lot for your Video lectures. It really helps people who want to learn in the field of Data Science. Learned a lot form you . Great work ..
What's missing at the end after the confidence interval is produced is a set of predictions made using out-of-sample data, for the purpose of assessing the accuracy on data other than the data which the model was trained on. Let us evaluate what percentage of forecasts using out of sample data are within the confidence intervals predicted by models m1, m2, m3. Thank you!
Hi, The video is helpful. What if I want to predict the real value, not just the residue? Should I add back the trend and seasonal components to the predicted residue values?
Hi Mr Bedi, I am following your youtube channel and thank you very much for the invaluable videos. I was wondering whether you would like to extend time series part with ARIMAX models. That would be a great contribution to youtube I think, not just your channel! I couldn't find any good youtube video for that. Best regards
Hi, You mentioned in your previous videos about the stationarity of time series before we apply any of these methods. Is this something we would have to check prior to apply AR/ARMA or any other forecasting techniques? Thanks in advance.
# I am attempting to find the percentage of true values from testing data # that fall within the predicted confidence interval of the model. # This program is almost correct, but it does NOT correctly compute # random + seasonal + trend of the confidence interval of the prediction. data(AirPassengers) ap=AirPassengers #Partition data into 90% training, 10% testing, nonrandomly selected for now because it is easier for first attempt. #90th percentile index of time series ap = floor(.9*144) = 129 library(xts) # convert object type ts to xts because ts::window() is too difficult to use. xap
This is not Random forest, you cannot choose training data. In fact no training is involved. You have to use all of the data set. Time Series theory is almost similar to linear regression which does not allow you to omit any observed data. Please note that Random forest is statistically invalid technique and its results are unacceptable for drug trials for USDA. The US govt will not approve any drug trial results which use training methods such as Random Forests.
hello, I ve couple of questions.. 1.what if the shapiro test gives pvalue less than 0.05? what should be my next step? and 2.if ACF and PACF plots doesnt give even single significant lag, that means i cant use the lag variable to forecast? Does the above 2 situation mean that my data is not stationary and i should be using differenced time series data? Thanks in advance!
Hi,
Thanks a lot for your Video lectures. It really helps people who want to learn in the field of Data Science. Learned a lot form you . Great work ..
Thanx alot sir. Very much helpful
What's missing at the end after the confidence interval is produced is a set of predictions made using out-of-sample data, for the purpose of assessing the accuracy on data other than the data which the model was trained on. Let us evaluate what percentage of forecasts using out of sample data are within the confidence intervals predicted by models m1, m2, m3. Thank you!
Best video I've seen on TS Prediction, it has an excellent balance of theory and hands-on exercise step-by-step.
Thanks!
+daspacer I am glad it helped. Please spread the word.
In general you should choose the highest LL, but the lowest AIC.
Thanks!
+JB I am glad it helped. Please spread the word.
Hi,
The video is helpful. What if I want to predict the real value, not just the residue? Should I add back the trend and seasonal components to the predicted residue values?
Yes
Hi Mr Bedi,
I am following your youtube channel and thank you very much for the invaluable videos.
I was wondering whether you would like to extend time series part with ARIMAX models.
That would be a great contribution to youtube I think, not just your channel! I couldn't find any good youtube video for that.
Best regards
+Cem Tekeşin Thanks Cem. Please spread the word about the UA-cam channel. I am not sure when I will have time to add more videos.
Hi,
You mentioned in your previous videos about the stationarity of time series before we apply any of these methods. Is this something we would have to check prior to apply AR/ARMA or any other forecasting techniques? Thanks in advance.
Check Seasonality in the data which needs to be isolated via linear regression, you can then proceed to apply time series.
# I am attempting to find the percentage of true values from testing data
# that fall within the predicted confidence interval of the model.
# This program is almost correct, but it does NOT correctly compute
# random + seasonal + trend of the confidence interval of the prediction.
data(AirPassengers)
ap=AirPassengers
#Partition data into 90% training, 10% testing, nonrandomly selected for now because it is easier for first attempt.
#90th percentile index of time series ap = floor(.9*144) = 129
library(xts)
# convert object type ts to xts because ts::window() is too difficult to use.
xap
This is not Random forest, you cannot choose training data. In fact no training is involved. You have to use all of the data set. Time Series theory is almost similar to linear regression which does not allow you to omit any observed data. Please note that Random forest is statistically invalid technique and its results are unacceptable for drug trials for USDA. The US govt will not approve any drug trial results which use training methods such as Random Forests.
hello,
I ve couple of questions..
1.what if the shapiro test gives pvalue less than 0.05? what should be my next step?
and
2.if ACF and PACF plots doesnt give even single significant lag, that means i cant use the lag variable to forecast?
Does the above 2 situation mean that my data is not stationary and i should be using differenced time series data?
Thanks in advance!
hi! please fix the link that is posted in the video description. thanks.
Done