Search

Explainable artificial intelligence (XAI) has shed light on enormous applications by clarifying why neural models make specific decisions. However, it remains challenging to measure how sensitive XAI solutions are to the explanations of neural models. Although different evaluation metrics have been proposed to measure sensitivity, the main focus has been on the visual and textual data. There is insufficient attention devoted to the sensitivity metrics tailored for time series data. In this paper, we formulate several metrics, including max short-term sensitivity (MSS), max long-term sensitivity (MLS), average short-term sensitivity (ASS) and average long-term sensitivity (ALS), that target the sensitivity of XAI models with respect to the generated and real time series. Our hypothesis is that for close series with the same labels, we obtain similar explanations. We evaluate three XAI models, LIME, integrated gradient (IG), and SmoothGrad (SG), on CN-Waterfall, a deep convolutional network. This network is a highly accurate time series classifier in affect computing. Our experiments rely on data-, metric- and XAI hyperparameter- related settings on the WESAD and MAHNOB-HCI datasets. The results reveal that (i) IG and LIME provide a lower sensitivity scale than SG in all the metrics and settings, potentially due to the lower scale of important scores generated by IG and LIME, (ii) the XAI models show higher sensitivities for a smaller window of data, (iii) the sensitivities of XAI models fluctuate when the network parameters and data properties change, and (iv) the XAI models provide unstable sensitivities under different settings of hyperparameters.

Multivariate time series with missing data is ubiquitous when the streaming data is collected by sensors or any other recording instruments. For instance, the outdoor sensors gathering different meteorological variables may encounter low material sensitivity to specific situations, leading to incomplete information gathering. This is problematic in time series prediction with massive missingness and different missing rate of variables. Contribution addressing this problem on the regression task of meteorological datasets by employing Long Short-Term Memory (LSTM), capable of controlling the information flow with its memory unit, is still missing. In this paper, we propose a novel model called forward and backward variable-sensitive LSTM (FBVS-LSTM) consisting of two decay mechanisms and some informative data. The model inputs are mainly the missing indicator, time intervals of missingness in both forward and backward direction and missing rate of each variable. We employ this information to address the so-called missing not at random (MNAR) mechanism. Separately learning the features of each parameter, the model becomes adapted to deal with massive missingness. We conduct our experiment on three real-world datasets for the air pollution forecasting. The results demonstrate that our model performed well along with other LSTM-derivation models in terms of prediction accuracy.