Cost-Aware Early Classification of Time Series

Abstract

Abstract. In time series classification, two antagonist notions are at stake. On the one hand, in most cases, the sooner the time series is classified, the more rewarding. On the other hand, an early classification is more likely to be erroneous. Most of the early classification methods have been designed to take a decision as soon as su cient level of reliability is reached. However, in many applications, delaying the decision with no guarantee that the reliability threshold will be met in the future can be costly. Recently, a framework dedicated to optimizing a trade-o↵ between classification accuracy and the cost of delaying the decision was proposed, together with an algorithm that decides online the optimal time instant to classify an incoming time series. On top of this framework, we build in this paper two di↵erent early classification algorithms that optimize a trade-o↵ between decision accuracy and the cost of delaying the decision. These algorithms are non-myopic in the sense that, even when classification is delayed, they can provide an estimate of when the optimal classification time is likely to occur. Our experiments on real datasets demonstrate that the proposed approaches are more robust than existing methods.

Self-assembling manifolds in single-cell RNA sequencing data

“…Because the recurrent response is found only in the 300 min time point (the latest time point in the study) and comprises only ~8% of these cells, we primarily focused on clustering cells based on their initial dynamics. To do this, we used the tslearn (Tavenard, 2017) python package to group cells based on their NF-κB activity time series. Because these time series are quite noisy, we were conservative in labeling cells as having a prolonged initial response in an effort to avoid false positives.…”

Lane Departure Warning Mechanism of Limited False Alarm Rate Using Extreme Learning Residual Network and ϵ-Greedy LSTM

“…In the distracted state, the driver often adjusts the vehicle, causing the vehicle to respond seriously. Here, Classification Accuracy, Average Cost Loss [33,34] and Computation Time are used as indicators to assess ELR-Net performance. Average Cost Loss can be calculated by…”

“…where n indicates the number of training time series, | represents the misclassification probability that the time series is classified at time t, expresses the delay cost of classification decision at time t, is the penalty factor (predetermined constant). Here, Classification Accuracy, Average Cost Loss [33,34] and Computation Time are used as indicators to assess ELR-Net performance. Average Cost Loss can be calculated by…”

Early classification of time series using multi-objective optimization techniques

“…However, this requirement is quite strict, and it is possible that, on many application domains, users are willing to sacrifice a bit on accuracy in order to obtain earlier predictions. Some recent solutions [6,26,27,28,32] include parameters or mechanisms to tune the weight associated to the accuracy and the earliness, finding different trade-offs between these two objectives. However, all these methods combine both objectives in some manner, formulating the problem as a single-optimization problem.…”

“…As such, they are not designed to treat the two objectives equally and they tend strongly towards one of the objectives: accuracy. Some of them such as [6,26,27,28,32], include some user-defined parameters which somehow enable modifying this stiff trade-off between the two objectives, but these parameters are usually difficult to tune in advance, and, additionally, in order to obtain solutions with varying trade-offs, we must execute the algorithms more than once. Indeed, to the best of our knowledge, all the proposed solutions reduce the problem of early classification to a single-objective problem by combining the two objectives in some manner.…”