Plot 3-way anova for given variable

Hello,

I have at my disposal features matrix including 12 variable. I would like to perform anova analysis for 3 variables, namely sensor, condition, class for each remaining feature separately. After that I would like to plot obtained 3-way anova results (means with standard deviations) for condition and class for each device separately. How to prepare such separated plots for device presenting conditiion and class?

Regards

E

Answer 1

[Abhas]

Hi @Elzbieta,

To perform a three-way ANOVA in MATLAB and plot the results for each device separately, you can follow these steps.

1. Ensure your data is organized in a table or matrix where each row represents an observation, and columns represent your variables (sensor, condition, class, and features).
2. Use MATLAB's ''anovan" function to perform the ANOVA for each feature. You'll need to pass your independent variables ('sensor', 'condition', 'class') and your dependent variable (each feature) to the function.
3. Use MATLAB's plotting functions to visualize the means and standard deviations for the 'condition' and 'class' of each device.

Here's an example code that demonstrates these steps. We'll create a dataset with 3 sensors, 2 conditions, 3 classes, and 3 features:

% Example data setup
rng(0); % For reproducibility

% Parameters
numObservations = 100;
sensors = {'Sensor1', 'Sensor2', 'Sensor3'};
conditions = {'Condition1', 'Condition2'};
classes = {'Class1', 'Class2', 'Class3'};
features = {'Feature1', 'Feature2', 'Feature3'};

% Generate random data
dataTable = table;
dataTable.Sensor = sensors(randi(length(sensors), numObservations, 1))';
dataTable.Condition = conditions(randi(length(conditions), numObservations, 1))';
dataTable.Class = classes(randi(length(classes), numObservations, 1))';

% Simulate feature data
dataTable.Feature1 = randn(numObservations, 1) + 5;
dataTable.Feature2 = randn(numObservations, 1) + 10;
dataTable.Feature3 = randn(numObservations, 1) + 15;

% Perform ANOVA and Plotting
for i = 1:length(features)
    feature = features{i};
    
    % Prepare data for ANOVA
    y = dataTable.(feature);
    g1 = dataTable.Sensor;
    g2 = dataTable.Condition;
    g3 = dataTable.Class;
    
    % Perform three-way ANOVA
    [p, tbl, stats] = anovan(y, {g1, g2, g3}, 'model', 'interaction', ...
                             'varnames', {'Sensor', 'Condition', 'Class'});
    
    % Plotting results
    for j = 1:length(sensors)
        sensor = sensors{j};
        
        % Filter data for the current sensor
        sensorData = dataTable(strcmp(dataTable.Sensor, sensor), :);
        
        % Calculate means and standard deviations
        means = zeros(length(conditions), length(classes));
        stddevs = zeros(length(conditions), length(classes));
        
        for k = 1:length(conditions)
            for l = 1:length(classes)
                subset = sensorData(strcmp(sensorData.Condition, conditions{k}) & ...
                                    strcmp(sensorData.Class, classes{l}), :);
                means(k, l) = mean(subset.(feature));
                stddevs(k, l) = std(subset.(feature));
            end
        end
        
        % Plot
        figure;
        hold on;
        for k = 1:length(conditions)
            errorbar(1:length(classes), means(k, :), stddevs(k, :), '-o', ...
                     'DisplayName', ['Condition ' num2str(k)]);
        end
        hold off;
        
        title(['Feature: ' feature ' for Sensor: ' sensor]);
        xlabel('Class');
        ylabel('Mean ± SD');
        xticks(1:length(classes));
        xticklabels(classes);
        legend('show');
    end
end

This code runs correctly in MATLAB, generating plots for each sensor showing the means and standard deviations of the features for different conditions and classes.

You may refer to the below MathWorks documentation link to perform N-way ANOVA: https://www.mathworks.com/help/stats/anovan.html