What is a 'classification score' in machine learning?

I wanted to implement machine learning algorithms (SVM and KNN) for a biometric-based authentication system to classify whether a user is genuine or not.

I have the feature data.

What I understood, in order for me to evaluate the performance of an algorithm using the equal error rate (EER), I have to obtain the score for each instance using a classification algorithm.

I have read about classification algorithm from books and papers. I couldn't find information on this.

Could you explain what is a classification score? Could you suggest me any reading material for me to understand the concept?

By the way, I'm using MATLAB.

migrated from stackoverflow.comJan 29 at 12:49

This question came from our site for professional and enthusiast programmers.

• a classification score is any score or metric the algorithm is using (or the user has set) that is used in order to compute the performance of the classification. Ie how well it works and its predictive power.. Each instance of the data gets its own classification score based on algorithm and metric used – Nikos M. Jan 29 at 10:29
• Have you heard about proper scoring rules and in particular for the Brier score? – usεr11852 Jan 31 at 23:18

You can use some different metrics to measure the scores of your classification model.

Normally, we use a confusion matrix to determine the correct/incorrect evaluations, and use this information to calc the metrics. The true positive/true negative are the values your model predict correctly, and the false positive/negative are the opposite The accuracy, for example, is the # of correct evaluations / # of cases, in other words, (TP+TN)/TP+TN+FP+FN

There are other metrics like precision, recall and F1.

• The question isn't about measures of classification performance, but about classification scores. I quote from the question: "I have to obtain the score for each instance using a classification algorithm", this is not the case with any measure based on the confusion matrix. The comment by Nikos is going in the right direction, although a more elaborate formulation would be better. – deemel Jan 31 at 12:09
• You're correct. – ItsMeMario Jan 31 at 18:59

Your classifier assigns a label to unseen previously data, usually methods before assignment evaluate likelihood of correct label occurrence. You can measure how good it is in many different ways, i.e you can evaluate how many of labels was assigned correctly (its called 'accuracy') or measure how 'good' was returned probability (i.e, 'auc', 'rmse', 'cross-entropy'). Metric which you choose will be call score. Usually score is fixed and set before you start looking for the best solution of problem.

So finally score is something which allow you to score your solution, and compare performance of different approaches.