Reliable early classification of time series

Anderson, Hyrum S.; Parrish, Nathan; Tsukida, Kristi; Gupta, Manisha

doi:10.1109/icassp.2012.6288318

Cited by 8 publications

(12 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This paper significantly extends a prior conference paper [1], where we tackled the same problem but proposed a more conservative decision rule. Here, we propose a more optimal, but still computationally tractable, decision rule.…”

Section: Introductionmentioning

confidence: 71%

“…Rule (2) differs from (1) in that only one set A that contains at least τ measure of X must be checked. This rule is more conservative than (1) because it does not check all sets A, and thus (1) could be satisfied without (2) being satisfied (but not vice-versa).…”

Section: Incomplete Decision Rulesmentioning

confidence: 99%

“…1 The Gaussian assumption is often justified by a maximum entropy argument or a simplicity argument.…”

Section: Naive Bayes Constructions For Set Amentioning

confidence: 99%

See 2 more Smart Citations

Early time-series classification with reliability guarantee.

Gupta

Parrish

Anderson³

2012

Self Cite

View full text Add to dashboard Cite

We consider the early classification of (incomplete) time-series data given a complete timeseries training set. The early classification problem arises naturally when test sample data is collected over time, or when costs must be incurred to collect the data. For example, for missile defense, it is important to determine the target type long before it reaches its target. A practical goal is to assign a class label as soon as enough data is available to make a good decision. This objective is formalized through the notion of reliability-the probability that a label assigned to the early, incomplete data matches that assigned to the complete data, and we propose a method to classify incomplete data only if a user-specified reliability threshold is met. Our approach models the complete data as a random variable whose distribution is dependent on the current incomplete data and the training data. The method differs from standard strategies in that our focus is on determining the reliability of the early classification decision, not only the accuracy. Proposed methods are tested on a set of open-domain time-series datasets; where the goal is to classify the time-series as early as possible while still guaranteeing that the reliability threshold is met. 3 4

show abstract

Section: Introductionmentioning

confidence: 71%

Section: Incomplete Decision Rulesmentioning

confidence: 99%

See 1 more Smart Citation

Early time-series classification with reliability guarantee.

Gupta

Parrish

Anderson³

2012

Self Cite

View full text Add to dashboard Cite

show abstract

“…When the algorithms see a value, they are assuming that it is z-normalized based on other values that do not yet exist! As we noted above, ECGs are a favorite example for ETSC papers [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20]. In Fig.…”

Section: Peeking Into the Futurementioning

confidence: 99%

“…1, assuming that all exemplars are of the same length and at least approximately aligned in time [1]. Given data formatted in this way, the ETSC community has produced dozens of models that can predict the class of an incoming subsequence, after only seeing a fraction of the data [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20]. This sounds impressive, but as shown in Fig.…”

Section: Introductionmentioning

confidence: 99%

When is Early Classification of Time Series Meaningful

Wu¹,

Der²,

Keogh³

2021

IEEE Trans. Knowl. Data Eng.

View full text Add to dashboard Cite

Since its introduction two decades ago, there has been increasing interest in the problem of early classification of time series. This problem generalizes classic time series classification to ask if we can classify a time series subsequence with sufficient accuracy and confidence after seeing only some prefix of a target pattern. The idea is that the earlier classification would allow us to take immediate action, in a domain in which some practical interventions are possible. For example, that intervention might be sounding an alarm or applying the brakes in an automobile. In this work, we make a surprising claim. In spite of the fact that there are dozens of papers on early classification of time series, it is not clear that any of them could ever work in a real-world setting. The problem is not with the algorithms per se but with the vague and underspecified problem description. Essentially all algorithms make implicit and unwarranted assumptions about the problem that will ensure that they will be plagued by false positives and false negatives even if their results suggested that they could obtain near-perfect results. We will explain our findings with novel insights and experiments and offer recommendations to the community.

show abstract