Machine learning explanability method for the multi-label classification model
Kushal Singla, Subham Biswas · 2021
Multi label classification is the identification of the multiple labels for a sample. There are a number of problems where a single label cannot be assigned to the sample and multiple labels are applicable. To take an action based on the multi label classification model, the model needs to be explained. There are many model agnostic machine learning explainability methods. However, the effectiveness of such methods for multi label classification model is not evaluated and the cognitive load of using the existing algorithms is high. In this paper, we propose a method for multi label model explainability and compare it with the LIME and CXPlain algorithm. We propose a method to compute a single list of ranked features explaining the multi-label model local explanation. The ranked list can be used for a variety of purposes such as, to debug the model misclassification, which provides the explanation quality similar to LIME but the time taken for using the explanation in a cognitive task is significantly lower. We show the results of evaluating the proposed method for a cognitive task on a private dataset and the open source dataset of CMU movie summary dataset. We get the hamming score of 90.75% on the private dataset and an evaluation time of 18 minutes for the sample set, and the hamming score of 65.23% on the public dataset and an evaluation time of 30 minutes.