How to use Machine Learning Algorithms in classification for categorical problem?

I have a matrix with 100*100 data points. I need to apply ML for classification of (Yes, there is an event to be 1, or No, there is no event 0). In addirion, I should only label 7500 (as 1 or 0) (75%) for training and no adding 1 or 0 for the remainder 2500 (25%) for testing?

Which models I should try? If I need to do comparative study, which algorithms I should try?

[the cyclist]

I'm confused (and I think you are, too).

You have a 100*100 matrix. What exactly is your response variable (the variable you are trying to predict)? What are your explanatory variables (the variables used to predict the response variable)?

Let's take a smaller example, where you just have a 5x5:

M = [0 0 0 0 1;
     0 1 1 1 0;
     1 1 1 1 1;
     0 0 0 0 0;
     1 0 1 0 1];

What are you trying to predict?

[Mohamed]

I will try to explain it more. This matrix is like a map (100 in x and 100 in y). This map values represent physical meausrements related to subsurface features, so high values means there is an object/anomaly/buried feature, and low values represent no event/feature or object at this location, so I need to give high values 1 and low values 0. After that, I need to train 75% of this data and test on 25%, but I need to break it as four blocks (top left,bottom left, and bottom right for training and top right block for testing).

I will be available for adding more information.

If it works will help in predict the locations of such features. I know if images, CNN is the better, but at this stage, I need to do in tabular data.

I can generate many data at the same locations (i.e., different physical value matrices that could confirm the location of buried features, but now I need to try on one data.

[Mohamed]

@the cyclist

[the cyclist]

This is helpful information, but it is still not clear how to make this into a classification problem. Let's modify my small example:

M = [10 20 30 40 50;
     20 35 45 55 60;
     25 40 60 75 65;
     25 20 30 40 50;
     20  5 15 35 45];

There are two points that are "local minimum" points: The value 10 at location (1,1), and the value 5 at location (5,2).

I also have a local maximum: the value 75 at location (3,4).

Is the first step to find the local minima? (That is not a machine learning problem.)

[Mohamed]

In the attached image, you can see that at the location of p1 and p2 the values are high compared to the surroundings. The target is high related to the location of the objects and low no objects present.

I am available for more details @the cyclist

Answer 1

[Image Analyst]

In the 100x100 matrix, is one of the columns the response (ground truth or known true value), and the other 99 columns are the predictors? Or do you have another vector for the response values?

Have you tried the Classification Learner app on the Apps tab of the tool ribbon? You should. You can then try a wide variety of different classification methods and pick the best one and export code for it.

If you have any more questions, then attach your data and code to read it in with the paperclip icon after you read this:

TUTORIAL: How to ask a question (on Answers) and get a fast answer

[Mohamed]

@Image Analyst

it is 100*100 matrix (as data points in map 100 in x and 100 in y). each value is related to location objects. imagine that there are some objects buried in this map (100*100 to scale) and at (200,250) there is an object and at differnt location in x and y there is another object. The presence of these objects will be reflected in the data measured at each data points, so over the object directly, the value is high and away is very low. Therefore, I need to give the location of the ground tructh (object is present 1 and no object present in each location 0), then I need to train this categorical problem and test on other portions of the data.

Can you suggest which will be better random forest or SVM or what?

I will be happy to hear your recommendations.

[Image Analyst]

I agree with @the cyclist - you just need to examine the image to find the peak(s), and the object(s) will be at the peak location.

Answer 2

[the cyclist]

Based on your replies to my comments, this does not seem like a machine learning classification problem to me. It seems like a peak-finding problem.

Take a look at this question/answer from the MathWorks support team, about 2-dimensional peak-finding. Maybe it will help you.

[Mohamed]

@the cyclist

My target is to find the location in x, y for any map of physical measurements to detect features. This example is simple. Some difficult examples has may peaks trends intersects with each other.

Also, there are different maps for the same locations; one of them is at negative values, the object is located. So how can make the algorithm from these two data sets and predict the location of objects.