Problem in Navie Bayes theorem code

Dear all

I used the below code on 160x5 file and found it work correctly but now i change the file from 160x5 to 160x17 then it show error. Kindly look into it.

clear all, close all, clc

load('FeatureAngle000Musscle1.mat') % 160x17 size

X = FeaturesAngle000Muscle1(:,1:16);

Y = FeaturesAngle000Muscle1(:,17);

rng(1); % For reproducibility

Mdl2 = fitcnb(X,Y,...

'DistributionNames',{'normal','normal','kernel','kernel' 'normal','normal','kernel','kernel' 'normal','normal','kernel','kernel' 'normal','normal','kernel','kernel'},...

'ClassNames',{'HandGrip','HandOpen','HandRest','WristExtension','WristFlexion'});

Mdl2.DistributionParameters{1,2}

isLabels2 = resubPredict(Mdl2); % it generate the output/response of model

ConfusionMat2 = confusionchart(table2array(Y),isLabels2); % Y need to convert

showing error

Error using ClassificationNaiveBayes/fitNonMNDists (line 222)

A normal distribution cannot be fit for the combination of class HandGrip and predictor LD. The data has zero variance.

Error in ClassificationNaiveBayes (line 104)

this.DistributionParameters = fitNonMNDists(this);

Error in classreg.learning.FitTemplate/fit (line 258)

[varargout{1:nargout}] = this.MakeFitObject(X,Y,W,this.ModelParams,fitArgs{:});

Error in ClassificationNaiveBayes.fit (line 132)

this = fit(temp,X,Y);

Error in fitcnb (line 252)

this = ClassificationNaiveBayes.fit(X,Y,RemainingArgs{:});

Answer 1

[Hiro Yoshino]

'ClassNames',{'HandGrip','HandOpen','HandRest','WristExtension','WristFlexion'});

the number of class labels is 5 but I guess yours is 17?

[Ali Asghar]

In 1-16 Columns there are different features values of emg signals like RMS, VAR etc and in 17th column there is respective output like handopen, handclose etc

In 1-32 rows the 17th columns have Handopen

in 33-64 the the 17th columns have Handclose

etc

Answer 2

[Hiro Yoshino]

I guess one of the distributions does not match what it is, i.e., variance is zero. In this case, Gaussian distributions cannot be fit anyway...

[Ali Asghar]

What i can i do now, other than Gaussian...

[saurabh kumar]

The error is due to the wrong distribution in the data sample. It occurs when the variance between the feature value is repeatedly 0 and gaussian distribution will find itself helpless in order to draw a distribution diagram. As for example, look into the attached screen shot. A dataset in my work contained 0s in most of the columns and I was validating the features via Naive Bayes algorithm, I got the same error .....

"A normal distribution cannot be fit for the combination of class 1 and predictor x3. The data has zero variance." .

There are two ways to solve the problem

a) Change your classifier (Use knn or M-SVM instead)

b) Use the following code to reduce your variance

function [odata] = reducevariance(data)

%REDUCEVARIANCE Summary of this function goes here

% Detailed explanation goes here

[rows,cols]=size(data); % find size of the data

odata=[]; % create updated data matrix

colcount=0; % column counter for new data

for j=1:cols

current_col_values=data(:,j); % take current col values

total_samples=numel(current_col_values); % find total number of samples

f=find(current_col_values==0); % find 0s in total samples

percentageofzeros=(numel(f)/total_samples)*100; % find % of existance

if percentageofzeros>20 % if it is greater than 20% remove the col

else

colcount=colcount+1; % else add the col to updated data

odata(1:rows,colcount)=data(1:rows,j);

end

end

end

For any other query, mail me at [e-mail]