A Parameter-Free Approach for Mining Robust Sequential Classification Rules

Abstract

Sequential data is generated in many domains of science and technology. Although many studies have been carried out for sequence classification in the past decade, the problem is still a challenge; particularly for pattern-based methods. We identify two important issues related to pattern-based sequence classification which motivate the present work: the curse of parameter tuning and the instability of common interestingness measures. To alleviate these issues, we suggest a new approach and framework for mining sequential rule patterns for classification purpose. We introduce a space of rule pattern models and a prior distribution defined on this model space. From this model space, we define a Bayesian criterion for evaluating the interest of sequential patterns. We also develop a parameter-free algorithm to efficiently mine sequential patterns from the model space. Extensive experiments show that (i) the new criterion identifies interesting and robust patterns, (ii) the direct use of the mined rules as new features in a classification process demonstrates higher inductive performance than the state-of-the-art sequential pattern based classifiers.

Anytime discovery of a diverse set of patterns with Monte Carlo tree search

“…Thus, the probability to draw an uninteresting pattern is still high, and not all local optima may be drawn: There are no guaranties on the diversity of the result set. Recently, the sampling algorithm Misere has been proposed by [24,20,21]. Contrary to the sampling method of Moens and Boley, Misere does not require any probability distribution.…”

“…direct-freq-roll-out top-10-mean-reward (7) direct-freq-roll-out random-reward (8) large-freq-roll-out (jumpLength = 10) max-reward (9) large-freq-roll-out (jumpLength = 10) mean-reward (10) large-freq-roll-out (jumpLength = 10) top-2-mean-reward (11) large-freq-roll-out (jumpLength = 10) top-5-mean-reward (12) large-freq-roll-out (jumpLength = 10) top-10-mean-reward (13) large-freq-roll-out (jumpLength = 10) random-reward (14) large-freq-roll-out (jumpLength = 20) max-reward (15) large-freq-roll-out (jumpLength = 20) mean-reward (16) large-freq-roll-out (jumpLength = 20) top-2-mean-reward (17) large-freq-roll-out (jumpLength = 20) top-5-mean-reward (18) large-freq-roll-out (jumpLength = 20) top-10-mean-reward (19) large-freq-roll-out (jumpLength = 20) random-reward (20) large-freq-roll-out (jumpLength = 50) max-reward (21) large-freq-roll-out (jumpLength = 50) mean-reward (22) large-freq-roll-out (jumpLength = 50) top-2-mean-reward (23) large-freq-roll-out (jumpLength = 50) top-5-mean-reward (24) large-freq-roll-out (jumpLength = 50) top-10-mean-reward (25) large-freq-roll-out (jumpLength = 50) random-reward (26) large-freq-roll-out (jumpLength = 100) max-reward (27) large-freq-roll-out (jumpLength = 100) mean-reward (28) large-freq-roll-out (jumpLength = 100) top-2-mean-reward (29) large-freq-roll-out (jumpLength = 100) top-5-mean-reward (30) large-freq-roll-out (jumpLength = 100) top-10-mean-reward (31) large-freq-roll-out (jumpLength = 100) random-reward room dataset. Finally, the description length is not or almost not influenced by the Roll-Out strategies.…”

Behavioral Constraint Template-Based Sequence Classification

“…In [17], the sequence database is split up in smaller parts to be recreated by a sparse knowledge base that punishes for infrequent behavior by constructing a Bayesian network of posteriors that are able to reconstruct the sequence database. A similar approach is used in [18], where a strong emphasis is used towards finding interesting sequences. In contrast to the previously mentioned techniques, iBCM draws from insights in constraint programming, but rather than constructing a complete constraint base that is able to elicit the sequence database as a whole, highly diverse and informative behavioral patterns are used that incorporate cardinality, alteration, gaps, as well as negative information.…”

“…Approaches iBCM is benchmarked against 4 other state-of-the-art techniques, being cSPADE [25], Interesting Sequence Miner (ISM) [17], Sequence Classification based on Interesting Sequences (SCIP) [15], and Mining Sequential Classification Rules (MiSeRe) [18], which all have the clear goal of obtaining discriminative, informative sequences for classification and are compared in Table 5. A comparison with other techniques can be found in the respective works as well.…”

“…max # optimization steps, max # iterations MiSeRe [18] Randomly generates diverse sequences and applies a Bayesian approach to retain interesting sequences.…”