In linear regression, the effect of each independent variables can be measured (has effect or not by looking at the significant value) and we can also determine if it has a positive or negative effect. Could such be done in ANN?
A bunch of thanks! The presentation is crystal clear! I have one question though. Conceptually, what does the hidden layer mean? Is it some sort of reference to mediating variables? Should I interpret the nodes as possible mediating variables that represent the mechanisms through which the IVs predict the DVs?
Thanks, Otmane. No, the hidden layers are not like mediating variables. They don't even represent real variables. Their only purpose is to enable a replication of the kind of circuitry that "works" in a human brain. You can vary hidden layers and number of nodes randomly.
Great presentation! I have a question. I have noticed something quite odd. When I create a neural network I get a certain output. But when i run it again, with exactly the same settings, the output (Importance, Normalized Importance) all change - why? How do i know which model is the best, given that "Percent Incorrect Predictions" always is zero or close to it?
Good question. Each time you run an algorithm, it gives slightly different results. You can manually bootstrap your model and then take an average of your predictions. That's what I do.
also in start you divide the data into training and testing sets, so it randomly takes data, and every time you run again the algorithm runs on different data thats because it takes data randomly for training and testing, but mostly the varying persentage is between 1-2 %
Artificial neural networks make no assumptions about the shape of the distributions. You can employ categorical variables as long as they are dummies (binary).
Using IQ for hiring decisions is highly problematic in terms of cost, applicant experience, racial adverse impact, and corporate bad-will. The decision to be made is hire or not (1,0) so the DV (appraisal rating) should be recoded to 1 or 0 based on what the organization calls acceptable perfoemcne (in this case perhaps > 7). In this way a wider net is cast and you can select from among a broader set of employees to hire (taking into consideration, perhaps, diversity/inclusion). In this scenario, you will find your predictions to be closer to 90% accurate because you now no longer call a predicted 7 (who was really an 8), a error. This is all sub-optimal mathematically, I know, but socially and legally more practicable in the world today.
This social justice troll sees your video as a generative theme to lecture you on antiracism and such and is best ignored.. 😂 even though it may be debatable whether there is much difference between 9 and 10 and 1 and 2 which were most mix-ups of the first model..
Several people have asked that I make these data available. You can find them on on Open Science Framework webpage: osf.io/drhw8/. Good luck with your predictions!
Thank youso much. How can we use the obtained prediction model with Neural Networks Using SPSS afterwards?
Good day. Can you share the data set just for our outline/basis regarding to our concept paper about neural networks. Thanks.
This is a helpful lecture! Thank you so much ครับ.
In linear regression, the effect of each independent variables can be measured (has effect or not by looking at the significant value) and we can also determine if it has a positive or negative effect. Could such be done in ANN?
Great! Thank you for sharing a good presentation.
please how can I know which cases take as training and which as testing?
Could you please guide me on how to do the model deployment in SPSS...
Dear
Can you share the data. the link you posted is inaccessible
Thank you so much for posting the helpful lecture!!!!
Great demonstration! Thank you
Wonderful explanation. Thank you so much
A bunch of thanks!
The presentation is crystal clear!
I have one question though. Conceptually, what does the hidden layer mean? Is it some sort of reference to mediating variables? Should I interpret the nodes as possible mediating variables that represent the mechanisms through which the IVs predict the DVs?
Thanks, Otmane. No, the hidden layers are not like mediating variables. They don't even represent real variables. Their only purpose is to enable a replication of the kind of circuitry that "works" in a human brain. You can vary hidden layers and number of nodes randomly.
@@andrewr.timming1931 Thank you for the time and the kindness!
Great presentation! I have a question. I have noticed something quite odd. When I create a neural network I get a certain output. But when i run it again, with exactly the same settings, the output (Importance, Normalized Importance) all change - why? How do i know which model is the best, given that "Percent Incorrect Predictions" always is zero or close to it?
Good question. Each time you run an algorithm, it gives slightly different results. You can manually bootstrap your model and then take an average of your predictions. That's what I do.
also in start you divide the data into training and testing sets, so it randomly takes data, and every time you run again the algorithm runs on different data thats because it takes data randomly for training and testing,
but mostly the varying persentage is between 1-2 %
I wonder why you mixed categorical and continuous variables together to run the model.
Artificial neural networks make no assumptions about the shape of the distributions. You can employ categorical variables as long as they are dummies (binary).
@@andrewr.timming1931 Thank you much for your response, much appreciated. .
Thanks for great tutorial
Hi,
Can I check what is the SPSS version that you used in this video?
Thanks.
I believe it was version 26. Shouldn't make a huge difference, though. The neural network modules haven't changed much.
Thank you so much.This video really helped me alot.Waiting for more videos on AI using SPSS,please post the same.
Awesome
Using IQ for hiring decisions is highly problematic in terms of cost, applicant experience, racial adverse impact, and corporate bad-will. The decision to be made is hire or not (1,0) so the DV (appraisal rating) should be recoded to 1 or 0 based on what the organization calls acceptable perfoemcne (in this case perhaps > 7). In this way a wider net is cast and you can select from among a broader set of employees to hire (taking into consideration, perhaps, diversity/inclusion). In this scenario, you will find your predictions to be closer to 90% accurate because you now no longer call a predicted 7 (who was really an 8), a error. This is all sub-optimal mathematically, I know, but socially and legally more practicable in the world today.
Umm, all data used in this exercise are hypothetical. The entire dataset was made up.
This social justice troll sees your video as a generative theme to lecture you on antiracism and such and is best ignored.. 😂 even though it may be debatable whether there is much difference between 9 and 10 and 1 and 2 which were most mix-ups of the first model..
Hello, was wondering if I can get the data, to practice with
No problem. I've included a link in the chat.
Several people have asked that I make these data available. You can find them on on Open Science Framework webpage: osf.io/drhw8/. Good luck with your predictions!
Not accessible
Thanks for great tutorial