1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Pattern discovery algoritms for weather prediction problem
Authors:	Yahyia Mohammed M. Ali Benyahmed (P67404)
Supervisor:	Azuraliza Abu Bakar, Prof. Dr.
Keywords:	Time-series analysis Weather -- Forecasting Data mining Universiti Kebangsaan Malaysia -- Dissertations
Issue Date:	17-May-2018
Description:	Pattern discovery in weather mining are performed to predict future events based on meaningful patterns extraction from previously observed events. The pattern is describing the behavior present at a specific time that required being useful and understandable for superior performance of prediction. Various algorithms are performed to process operational prediction based on pattern discovery. The tasks are complex and difficult since the weather data is collected events over time for unusual and surprised phenomena. The weather data problem is the high-dimensional data, which are dealing in its raw format is memory consuming, noisy, complexity searching and huge events. It is needed to develop representation techniques that can reduce the dimensionality of data without substantial loss of information. The aim of study presents an effective weather pattern discovery algorithm which includes the processes of relevant and interesting pattern, to improve the effectiveness of weather prediction, with three main objectives: i) time series representation; ii) prediction; iii) pattern discovery. The first is to enhance Symbolic Aggregate approXimation (SAX) algorithm for weather representation called ESAX+ that can reduce the dimensionality with less loss information. The second is to propose predictor model based on Naive Bayesian algorithm (NB) for prediction pattern that integrated with ESAX+ representation and sliding windows approach to extract the most suitable patterns to find subsequence patterns of weather data. The third is to enhance dynamic pattern detection approach using a sliding window algorithm for weather data segmentation called ESW+, which is using change-point detection proposed to extract the meaningful patterns from weather data that influence to weather changes and pattern discovery algorithm applied for the frequent pattern and sequential patterns of weather data. The performance of the proposed algorithm was evaluated using UCR time series data, and by the Malaysian weather data for rainfall and river flow applications, which were collected from 13 stations for 35 years period. The proposed solutions achieve encouraging performance of pattern prediction and pattern discovery that were supported and trusted from the experts. Experimental results show that the proposed symbolic representation has superior performance in several existing algorithms. The ESAX+ Algorithm shows better results as average as 0.426 in terms of error rate based on optimal word and alphabet size of time series data, the algorithm improved the SAX algorithm. In pattern prediction, NB approach was able to generate significant patterns and rules for pattern prediction with superior prediction accuracy up to 79% and was supported by the experts. In pattern discovery, the proposed combined mining algorithms were able to generate higher confidence as high as 70% and, minimum support 0.01 for frequent and sequential patterns. The proposed study has shown its potential in generating algorithms that have the ability to maintain vital knowledge and reduce information loss. Therefore, the weather prediction task has exposed more essential information that can support the decision-making process.,Ph.D.
Pages:	271
Call Number:	QA280.B469 2018 3 tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.