Pusat Pengajian Sains Matematik - Tesis
Browse
Recent Submissions
- PublicationBayesian Networks With Greedy Backward Elimination In Feature Selection For Data Classification(2019-03)Ang, Sau LoongNaive Bayes (NB) is an efficient Bayesian classifier with wide range of applications in data classification. Having the advantage with its simple structure. Naive Bayes gains attention among the researchers with its good accuracy in classification result. Nevertheless, the major drawback of Naive Bayes is the strong independence assumption among the features which is restrictive. This weakness causes not only confusion in the causal relationships among the features but also doubtful representation of the real structure of Bayesian Network for classification. Further development of Naive Bayes in augmenting extra links or dependent relationships between the features such as the Tree Augmented Naive Bayes (TAN) end up with slight improvement in accuracy of classification result where the main problems stated above remain unsolved.
- PublicationNumerical Solutions For Two Dimensional Time-Fractional Differential Sub-Diffusion Equation(2019-04)Ali, UmairIn the past several decades, fractional differential equations (differential equation involving arbitrary order derivatives) have acquired much popularity in the area of science and engineering. This is because such equations can better model certain problems of fluid mechanics, physics, biological science, chemistry, hydrology and finance, amongst others, due to the fact that it can better represent system with memory. However, most fractional differential equations cannot be solved by exact analytical techniques.
- PublicationEfficient Entropy-Based Decoding Algorithms For Higher-Order Hidden Markov Model(2019-03)Chan, Chin TiongHigher-order Hidden Markov model (HHMM) has a higher prediction accuracy than the first-order Hidden Markov model (HMM). This is due to more exploration of the historical state information for predicting the next state found in HHMM. State sequence for HHMM is invisible but the classical Viterbi algorithm is able to track the optimal state sequence. The extended entropy-based Viterbi algorithm is proposed for decoding HHMM. This algorithm is a memory-efficient algorithm due to its required memory space that is time independent. In other words, the required memory is not subjected to the length of the observational sequence. The entropybased Viterbi algorithm with a reduction approach (EVRA) is also introduced for decoding HHMM. The required memory of this algorithm is also time independent. In addition, the optimal state sequence obtained by the EVRA algorithm is the same as that obtained by the classical Viterbi algorithm for HHMM.
- PublicationArrangement Of Letters In Words Using Parikh Matrices(2019-04)Poovanandran, GhajendranThe Parikh matrix mapping is an ingenious generalization of the classical Parikh mapping in the aim to arithmetize words by numbers. Two words are M-equivalent if and only if they share the same Parikh matrix. The characterization of M-equivalent words remains open even for the case of the ternary alphabet. Due to the dependency of Parikh matrices on the ordering of the alphabet, the notion of strong M-equivalence was proposed as an order-independent alternative to M-equivalence. In this work, we introduce a new symmetric transformation that justifies strong M-equivalence for the ternary alphabet. We then extend certain work of §erbanuja to the context of strong ^-equivalence and show that the number of strongly M-unambiguous prints for any alphabet is always finite.
- PublicationInterpolasi Data Berselerak(1994-03)Rahmita WirzaTesis ini bertujuan untuk meninjau teknik-teknik bagi memodelkan permukaan yang licin bagi data berselerak dan penganggaran data berselerak tiga dimensi yang diberikan. Kaedah yang akan dibincangkan melibatkan tiga langkah iaitu, penyegitigaan data domain; penganggaran terbitan separa pada setiap data yang diberi untuk menghasilkan permukaan yang licin dan penakrifan skema interpolasi.