CODEWITHSUNDEEP
machine-learningwekaroc
dkea
dkea

Reputation: 708

What is the best measure for ACCURACY in MACHINE LEARNING

I have been researching about Machine Learning concepts but I still have not been able to properly grasp which attribute I can use to determine what could tell the true accuracy of my system (if its the TPR or the Correctly Classified Instances)

I am also looking for a good explanation of ROC and Kappa Statistic. As far as I know, ROC measures how much the system is learning and Kappa measures how much the system is guessing. However, those are the explanations of our professor who's had years of experience. Looking at other technical documents, we could not figure out what he means by those terms.

So, I just need to know what could tell the accuracy of my algorithm and what attributes are needed to support if the algorithm is good such as the ROC and Kappa Statistic.

Please help! Thank you very much!

Upvotes: 2

Views: 1984

Answers (1)

fatihk
fatihk

Reputation: 7929

Two basic criteria about the machine learning performance is false positive(wrong alarm) ratio and false negative (missed target) ratio. Lets say, you are looking for a particular object type and if you detect 10 wrong objects as target out of 100 out-class objects then false positive ratio is 10%.

if you label 10 true objects as out-class out of 100 target objects, then false negative raio is 10%.

Usually there is a trade-off between this two ratios and area of ROC curve is a good indicator of performance

Upvotes: 1

Related Questions