I am trying understand Artificial Intelligence Neural Network and I am self-learner. Hope anyone would help me in understanding on how to solve this problem
If this post should be posted here. Please comment instead of degrading the post. Appreciate for this as well.
I have a question that I am totally confused about how to solve it. I encountered this online but was unable to understand how to solve it. I have added the question below. Hope you can provide some help.
The data set contains 4 observations for 4 input variables (Temp, Pres, Flow, and Process) and an output variable (Rejects). The first column "No" is simply an identifier. The table below reproduces the first 4 observations:
| No | Temp | Pres | Flow | Process | Rejects |
|---|---|---|---|---|---|
| 1 | 53.39 | 10.52 | 4.82 | 0 | 1.88 |
| 2 | 46.23 | 15.13 | 5.31 | 0 | 2.13 |
| 3 | 42.85 | 18.79 | 3.59 | 0 | 2.66 |
| 4 | 53.09 | 18.33 | 3.67 | 0 | 2.03 |
Train a back-propagation neural network on approximately 80% of the observations, randomly selected. Test the trained network using the remaining 20% observations.
Question:
- Based on this how to define a fixed neural network with output values and backpropagate an expected output pattern? Here, the output is only one which is the "Rejects" Column
- What are the error values which is required to be calculated?
- Does it required to define the hidden layer here? And how can we define the hidden layer?
- What type of “tool” can be used to create a report for the above inputs and get the expected output? Can you help related to this? I am unsure about one thing as well
- If not tool could you provide any program to understand this? Preferable tool though.
- Create a figure that plots the actual and predicted values of the output "Rejects" for the training and test data sets.
- Does this mean creating a chart something similar to the plot chart we create for the Support Vector Machine? Is that possible to create in the tool where we are using for the above question?
- How to solve -> Sum of squared errors for the training and test data sets.
I would really appreciate your help.
2
Answers
Firstly, the dataset is insanely small. However, this is the way you would approach this kind of dataset, assuming there is much more data.
The results are not good, but that is only due to the quantity of data.
So I’m writing a fresh answer because the code for this is tuned to suit the dataset URL that you have provided after the first answer. This time clearly the accuracy is much better due to the quantity of data available.
And the model training went something like this: