Mining Fuzzy Sequential Patterns in Multi-items and Multi-quantities Transaction Database

• Main Authors: WANG, KUAN-LUNG, 王觀隆
• Other Authors: 王乃堅
• Format: Others
• Language: en_US
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/22838303542754847814

碩士 === 國立臺灣科技大學 === 電機工程系 === 89 === The problem of mining fuzzy sequential patterns based on a given transaction database is proposed in this thesis. Each customer transaction in database includes customer-ID, transaction-time, and the items with its quantitative value. We present three improved algorithms to overcome the problem with multi-items and multi-quantities characteristic. The support counting method combines the fuzzy concepts to calculate the support value for candidate fuzzy itemsets and candidate fuzzy sequences. The experimental results are obtained by change parameters and user predefined minimum support values in litemset and fuzzy sequence phases. FuzzyAprioriSome works better than FuzzyAprioriAll when min_support is constant in fuzzy sequence phase and the other min_support in litemset phase decreases. On the contrary, FuzzyAprioriAll works better than FuzzyAprioriSome when min_support is constant in litemset phase and the other min_support in fuzzy sequence phase decreases. But FuzzyDynamicSome is the worst one among three algorithms no matter in what cases.