I want to run a ML regression model for my variables which are in format of map( image). Do I need to quantify the map to run the model?

Hi,

I am new in MATLAB and for my research, I started to learn how to run machine learning based models in MATLAB.

Here is a bit about my project:

The dependent variable (Mean Radiant Temperature) and the independent variables(Urban Spatial Factors: building height, H/W, LULC,... some are numeric and others are ) are all maps(output of other application). What I am expecting from this model is to predict MRT based on what different value of other dependent variables.

My question is do I need to quantify the maps before importing into matlab? and How can I do that?

Answer 1

[Udit06]

Hi Modjrian,

I understand that you want to use machine learning models to predict mean radiant temperature(MRT) and your independent variables include image maps obtained as output of other applications. It is not necessary to quantify the maps before importing them for machine learning tasks, but you could apply some preprocessing steps on the image data before using it for training machine learning model.

1) If your maps have different statistical distributions, you can normalize them to have zero mean and unit variance using the "zscore" function. You can refer the following MathWorks documentation to understand more about the "zscore" function.

https://www.mathworks.com/help/stats/zscore.html

2) If your maps have multiple channels or layers, you may want to extract specific features or combine them into a single representation. You can refer to the following MathWorks documentation to understand more about different feature extraction techniques.

https://www.mathworks.com/help/vision/feature-detection-and-extraction.html

I hope this helps.