000			03611ntm a2200337 i 4500
003			MY-KuUP
005			20260220161458.0
006			t||||fr|||| 000 0
007			ta
008			260220t20252025my a|||fr|||| 000 0 eng d
020			_aTHE0010361 (Local) _qHardback
040			_aUMPSA _beng _cUMPSA _erda
090			_aFKOM .E37 2025 r Thesis
100	1		_aEkramul Haque Tusher, _eauthor.
245	1	0	_aA hybrid approach to fuzziness based long short-term memory network for email spam detection / _cEkramul Haque Tusher
264		1	_aKuantan, Pahang : _bUMPSA, _c2025
264		4	_c© 2025
300			_axiii, 119 pages : _billustrations ; _c30 cm. + _e1 CD-ROM
336			_2rdacontent _atext
337			_2rdamedia _aunmediated
338			_2rdacarrier _avolume
347			_2rda _atext file _bPDF
500			_aFaculty of Computing
502			_aThesis (Doctor of Philosophy) -- Universiti Malaysia Pahang Al Sultan Abdullah - 2025
504			_aInclude bibliographical reference
520	3		_aSpam email detection is a critical task in cybersecurity, as spam messages often contain phishing attempts, malware, or fraudulent content. Traditional rule-based systems and machine learning (ML) methods struggle to adapt to evolving spam tactics, leading to decreased detection accuracy. To address this challenge, deep learning (DL) methods have been widely explored due to their ability to model complex data patterns effectively. However, while fuzzy systems offer interpretability and adaptability in decision-making, they often fail to classify complex datasets with high accuracy. Similarly, Long Short-Term Memory (LSTM) networks, known for their ability to process sequential data, face challenges related to memorizing noise, which limits their effectiveness in spam classification. This highlights the need for a robust approach that combines the interpretability of fuzzy systems with the sequence modeling capabilities of LSTM network to improve spam detection accuracy while effectively managing uncertainty. To overcome these limitations, this study introduces a hybrid approach called Fuzziness-based LSTM for Email Spam Detection (FLSTM-ESD), which integrates fuzzy logic with LSTM networks. A pilot study was conducted to compare the performance of four DL methods: Artificial Neural Network (ANN), LSTM, Gated Recurrent Unit (GRU), and Bidirectional LSTM (Bi-LSTM). The results identified LSTM as the most effective model for detecting sophisticated spam patterns. Based on this finding, a fuzziness-based divide-and-conquer strategy was implemented to enhance classifier performance by incorporating low-fuzziness samples into the training process, which was experimentally confirmed to improve accuracy. Experimental results demonstrated that FLSTM-ESD outperformed other DL methods, achieving the highest classification accuracy across all tested datasets. The method achieved 94.9% accuracy on the Spambase dataset, 96.84% on the Spamassassin dataset, and 97.09% accuracy on the LingSpam dataset. Compared to previous studies, FLSTM-ESD establishes itself as a state-of-the-art method for email spam detection, surpassing existing approaches. The integration of fuzzy logic with LSTM improves spam classification performance while effectively handling uncertainty and evolving spam patterns. These findings confirm that FLSTM-ESD is a robust and reliable approach for real-world email spam detection applications.
610	2	0	_aFaculty of Computing _xDissertations
650		0	_aUniversities and colleges _xDissertations
650		0	_aTheses _xDissertations
942			_2lcc _cTHESIS
999			_c104214 _d104220