Sequence Classification Based on Delta-Free Sequential Patterns

Abstract

Sequential pattern mining is one of the most studied and challenging tasks in data mining. However, the extension of well-known methods from many other classical patterns to sequences is not a trivial task. In this paper we study the notion of δ-freeness for sequences. While this notion has extensively been discussed for itemsets, this work is the first to extend it to sequences. We define an efficient algorithm devoted to the extraction of δ-free sequential patterns. Furthermore, we show the advantage of the δ-free sequences and highlight their importance when building sequence classifiers, and we show how they can be used to address the feature selection problem in statistical classifiers, as well as to build symbolic classifiers which optimizes both accuracy and earliness of predictions.

Reference-Based Sequence Classification

“…The selected pattern should satisfy the following criteria: (1) be frequent, (2) be distinctive of at least one class and (3) not redundant. Towards this direction, many pattern-based classification methods have been subsequently proposed, in which different constraints are imposed on the patterns that should be selected as features [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [26], [27], [28], [29]. Note that any classifier designed for vectorial data can be applied to the transformed data generated from such pattern-based methods.…”

“…frequent pattern, discriminative pattern). Over the past few decades, a large number of pattern-based methods have been presented in the context of sequence classification [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29].…”

A Parameter-Free Approach for Mining Robust Sequential Classification Rules

“…This section presents a comparative study of the performance of MiSeRe and state-of-the-art competitive rule mining algorithms with several classification methods. We compare the set of rules mined by MiSeRe with four baseline algorithms: (1) cSPADE [27], (2) SCII [9], (3) Gokrimp [32], and (4) DeFFeD [10]. The parameters were set for each algorithm as indicated in the original papers.…”

“…Generally, pattern-based classification methods [7] follow a similar strategy: firstly, a sequential rule set is mined w.r.t. an interestingness measure; secondly, either a dedicated classifier, like a decision list or a Maximum Entropy model, is built upon a selected subset of the mined rules [8], [9], [10] or the mined rules are directly used as new features in a classification process [11], [12], [13]. While most of the existing approaches generally lead to good inductive performance, we now highlight two of their weaknesses, namely the curse of parameter tuning and the instability of the interestingness measures.…”

“…MiSeRe input : D, a Labeled Sequential Data Set output: R, a Set of SCRMs 1 begin ={s = s1 ; s1 ∈ I} ; ={π : s → fc 1 (s), · · · , fc j (s); s ∈ S ∧ level(π) > 0} ; ChooseRandomSequence(D) ; ComputeNumberOfSubsequences(s) ;7 for i = 1 to log(ads) do = GenerateRandomSubsequence(s) ; : s → fc 1 (s ), · · · , fc j (s ) ;10 if level(π) > 0 ∧ π ∈ R then 11 R = R ∪ {π} 12 return R;…”