Theoretically , one class svm is not that different from the usual SVM as it tries to find the optimal hyperplane that separates data inliers ( data that have a certain pattern linking them = gaussian kernel phi ( x,x' ) ~ 1 ) from the outliers , so if you're deciding to use a gaussian kernel , you can have your anomaly score as the distance of the point from the origin the high dimensional space which is nothing more than its norm , thus the lower it is , the more likely the point is an outlier as SVM tries to maximize the distance separating the hyperplane from the origin. ( Same thing but in another perspective , you can have your anomaly score as the distance separating the point from the hyperplane , the bigger it is , the more likely the point is an outlier )
The screenshot i uploaded is from an article i read once during my internship , here is the link : https://www.analyticsvidhya.com/blog/2024/03/one-class-svm-for-anomaly-detection/.

Good luck :) enter image description here