I’m a soil scientist and I think a machine learning approach would be a good solution for making predictions from my data sets. I’m trying to set up a neural network to make predictions about soil properties based on measured data. I’m new to machine learning so I’m reaching out to the data science community.
An individual input data set is two dimensional consisting of two columns and 1200 rows. The data itself is a voltage measurement along a transition line at 1200 points that are evenly spaced. The output that I’m looking for is basically a single capacitance value from 1 to 80 for each input data set. I have about 20 to 40 data sets of known outcomes to use to train a model. The output will be a single number from 1 to 80 with a resolution of 0.1. Because I don’t have much training data, the model will need to interpolate between the reference training points that I have. The reason why I think a neural network is a good solution to this problem is because if you plot the input data, it looks like a hand written character. Also, just like with hand written letters, " you know it when you see it". The size of the hills and valleys in the plot will influence the final output. In other words, the distance between two hills in the plot can be quantified to calculate the desired output. The graph below is an example and has a capacitive value of 25.9.
I want to be able take the input data and predict the capacitance value.
I wrote code in python setting variables to the sizes of the hills and valleys in the input data to calculate the output, but it is not always accurate in some soil types because things get smushed. There are tons of examples online on how to set up a neural network to identify categorical type things such hand written letters or pictures of cats. But how do you setup a model to look at a squiggly line and predict number value? I think a can muddle through coding it in python and in R OK. I’ve used TensorFlow for other things. I would be grateful and appreciate any insight.
Thanks
