MFlexDT: multi flexible fuzzy decision tree for data stream classification

Abstract

In many real-world applications, instances (data) arrive sequentially in the form of streams. Processing such data poses challenges to machine learning. While adhering to on-line learning strategies, in this paper we extend the Flexible Fuzzy Decision Tree (FlexDT) algorithm with multiple partitioning that makes it possible to carry out automatic online fuzzy data classification. The proposed method is aimed to balance accuracy and tree size in data stream mining. The objective of the classification problem is to predict the true class of each incoming instances in real time. In terms of evaluation of the method, accuracy, tree depth, and the learning time are significant factors influencing the performance. A series of experiments demonstrate that the proposed method produces optimal trees for both numeric and nominal features (variables).

Incremental Learning of Fuzzy Decision Trees for Streaming Data Classification

“…Both FlexDT and Multi FlexDT generate, incrementally, noise-robust models and also allow managing efficiently missing values. As stated in [10], Multi FlexDT outperforms FlexDT in terms of accuracy and depth of the trees. However, linguistic labels cannot be defined a-priori for each input attribute and a good level of integrity of the fuzzy partitions is not ensured.…”

“…The classical formulations of the Hoeffding bound and of the information gain are used, respectively, for deciding whether to expand or not the tree and to select the best input attribute to be used for the splitting at each node. Recently, an extension of FlexDT, denoted as Multi FlexDT, has been introduced in [10]. In Multi FlexDT multi-way splits rather than binary splits are allowed at each decision node.…”

Aplicando método do gradiente ótimo na otimização do cálculo do grau de cobertura das regras em árvores de decisão Fuzzy

“…[10] e [11] criou uma versão de um algoritmo de arvore de decisão fuzzy chamada FuzzyDT, que mostrou-se uma versão com baixa taxa de erros, sendo escolhida como ponto de partida para este trabalho devido sua simplicidade. Foram ainda considerados outros métodos que lidam com algoritmos de arvore de decisão fuzzy como em [3], [12] e [13].…”

“…O valor de α j é determinado de maneira similar a seleção do ponto de corte encontrado nas arvores de decisão tradicionalmente binária [13] .Para cada conjunto de instancias fuzzy S e característica x j , a informação de entropia paras as partições direita e esquerda(S R , S L ) é estimado para qualquer provável ponto de corte α da seguinte maneira em (4):…”

Decision Trees based Fuzzy Rules

“…They are also recognized as highly unstable classifiers with respect to minor perturbations in the training data, in other words, methods presenting high variance. Fuzzy sets and fuzzy arithmetic provide a vehicle to use linguistic and qualitative features to form rules that are more understandable and accurate [7].…”