Pusat Pengajian Sains Matematik - Tesis

Browse

Now showing 1 - 5 of 474

Hybridizing Structural Time Series With Dynamic Nonlinear Autoregressive Neural Networks To Improve Cryptocurrency Price Prediction
(2025-03)
Rashid, Nurazlina Abdul
This study proposes an alternative hybrid model that combines structural time series (sts) with dynamic nonlinear autoregressive neural networks (nar and narx) to improve cryptocurrency price predictions. Given that cryptocurrency market movements are complex, caused by volatility, nonstationarity, and nonlinearity, as well as being influenced by hidden and external factors, accurate forecasting presents a challenge, as traditional models often fail to capture the market's complex dynamics. Addressing these limitations, this research integrates sts models, renowned for their ability to handle nonstationarity and modeling hidden factors, including trends, seasonality, irregular components, and external factors, with dynamic nonlinear autoregressive neural network models capable of managing nonlinear patterns. Utilizing a detailed analysis of the top five cryptocurrencies by their market capitalization as of december 2022, the study adopts a multi-stage methodology in order to achieve the research objectives. It begins with identifying initial trend behaviors through regression model, followed by modeling hidden factors using sts. The innovation of this study lies in enhancing these models with hybrid sts-narx and sts-nar frameworks, significantly improving predictive accuracy.
A New Generalized Trigonometric Bernstein-Like Basis Functions And Its Applications In Curve And Surface Constructions
(2025-03)
Ammad, Muhammad
This study presents a novel methodology for constructing curves and surfaces using the gt-bernstein-like basis function with two design variables. The proposed curves and surfaces substantially enhance shape-adjustment capabilities compared to traditional forms. The research explores the application of these curves and surfaces, enabling the creation of shape-adjustable surfaces with local control, such as swept surfaces, swung surfaces, rotation surfaces, ruled surfaces, enveloping surfaces, and spine curves of developable surfaces. A thorough analysis of different parameters shows the influence of the shape of these curves and surfaces, leading to the identification of optimized parameters for shape optimization design. Numerical examples showcase the flexibility and local shape control achieved through the proposed method. Additionally, the study advances a new method for generating surfaces with prescribed boundaries in computer-aided geometric design (cagd), with a specific focus on minimizing surface area. The
Negative Based Higher Order Systematic Satisfiability Logic With Hybrid Black Hole Algorithm In Enhancing Multiunit Discrete Hopfield Neural Network
(2025-06)
Rusdi, Nur ‘Afifah
Understanding intelligence is crucial for developing advanced intelligent models. In pursuit of this goal, satisfiability logical representation in Discrete Hopfield Neural Network has provided new insight in understanding the behaviour of the data. However, the role of negation in understanding intelligence has been overlooked as negation is often associated with false outcome. Negative Based Higher Order Systematic Satisfiability Logic is proposed to promote the appearance of negative literals within the clauses. The proposed logic demonstrates optimal performance as compared to existing logical rules. To further improve the quality of the final neuron states, Hybrid Black Hole Algorithm is proposed to update the neuron states that satisfy the multi-objective functions. The newly proposed mechanism will be incorporated into the logic mining model known as Multi-unit Negative Based Higher Order Systematic Satisfiability Reverse Analysis.
Hybrid Model Of Climate Effects On Dengue Forecast And Control In Selangor, Malaysia
(2025-07)
Lu, Xinyi
Existing dengue forecasting approaches in Malaysia are constrained by two main limitations: deterministic models typically assume a constant mosquito biting rate, neglecting its climate-driven variability, while statistical time series models are usually effective only for short-term forecasts. These constraints hinder the accurate prediction of outbreak dynamics and limit the utility of earlywarning systems. This study develops two integrated approaches: (i) coupled Susceptible-Infective for vector populations and Susceptible-Infective-Recovered for human populations (SI-SIR) and Autoregressive Integrated Moving Average with eXogenous variables (ARIMAX) model, and (ii) SISIR with Multiple Linear Regression and Long Short-Term Memory (MLR-LSTM) model.
Modified Iterative And Repetitive Estimation Methods For Fitting Measurement Error Model
(2025-03)
Al-Dibi’I, Ro’Ya Saleh Faleh
This thesis introduces new estimation methods for fitting structural measurement error model (mem), addressing challenges in estimating relationships where all variables are subject to error. The study focuses on distribution-free weighted estimation techniques, which improve parameter estimation without the strict assumptions required by traditional methods. A new weighted wald-type estimation procedure, developed in modified iterative and repetitive forms, is proposed and tested on both simple and multiple mems. Simulation studies show that these methods achieve lower mean square errors across various weighting scenarios, outperforming classical approaches such as maximum likelihood estimation (mle) and the method of moments (mom) in terms of accuracy and reliability. Additionally, robust weighted estimation methods, including theil and siegel procedures, demonstrate superior performance in different settings. The proposed techniques are applied to two real datasets: one examining the relationship between infant deaths, hepatitis b, and polio, revealing strong negative correlations, and another analyzing gross domestic product (gdp), unemployment, and the human development index (hdi) in jordan, highlighting strong positive correlations between gdp and hdi and a strong negative relationship between unemployment and hdi.

Browse

Recent Submissions