Publications

The following is my publications throughout my academia years so far...
(Updated: July 5, 2021)

View me on LinkedIn
View my Google Scholar
View me on ResearchGate

Predictive Analytics for Crude Oil Price Using RNN-LSTM Neural Network

Prediction of future crude oil price is considered a significant challenge due to the extremely complex, chaotic, and dynamic nature of the market and stakeholder's perception. The crude oil price changes every minute, and millions of shares ownerships are traded everyday. The market price for commodity such as crude oil is influenced by many factors including news, supply-and-demand gap, labour costs, amount of remaining resources, as well as stakeholders' perception. Therefore, various indicators for technical analysis have been utilized for the purpose of predicting the future crude oil price. Recently, many researchers have turned to machine learning approached to cater to this problem. This study demonstrated the use of RNN-LSTM networks for predicting the crude oil price based on historical data alongside other technical analysis indicators. This study aims to certify the capability of a prediction model built based on the RNN-LSTM network to predict the future price of crude oil. The developed model is trained and evaluated against accuracy matrices to assess the capability of the network to provide an improvement of the accuracy of crude oil price prediction as compared to other strategies. The result obtained from the model shows a promising prediction capability of the RNN-LSTM algorithm for predicting crude oil price movement. (Read full paper)

A Spiking Neural Networks Model with Fuzzy-Weighted k-Nearest Neighbour Classifier for Real-World Flood Risk Assessment

Inspired by the brain working mechanism, the spiking neural networks has proven the capability of revealing significant association between different variables spike behavior during an event. The combination of the capability of SNN to produce personalised model has allowed high-precision for data classification. The exiting accuracy of weighted k-nearest neighbors classifier being used in the spiking neural networks architecture, noticeably can be further improved by implementing fuzzy-weights on the features, therefore allowing data to be classified more precisely to the high-impacting features. Simulation has been done by using three classifiers—Multi-layer Perceptron, weighted k-nearest neighbors, and Fuzzy-weighted k-nearest neighbors (FwkNN) using a real-world flood case study dataset and two benchmark dataset. Based on the result using the Kuala Krai Rainfall Dataset, FwkNN classifier has improved accuracy by 3.48% and 3.57% for 3-days earlier and 1-day earlier classification respectively. As compared to, FwkNN classifier has proven the capability to reduce misclassification and increase the accuracy of dataset classification. (Read full paper)

Improved Spatio-Temporal Data Modelling with Spiking Neural Network Methods for Environment Event Prediction

This research aims to: 1. Propose a new classifier to improve classification result of the output from a SNNr. 2. Develop a new classifier that combines SNN methods and personalised modelling methods for improving the result of SSTD data classification and analysing regression problems. 3. Validate the algorithm via supervised learning experiment on real-world flood event case study data by comparing the accuracy of prediction using the wkNN classifier with the proposed wwkNN classifier. (Read full paper)

Comparative Analysis of Spatio/Spectro-Temporal Data Modelling Techniques

Recent advancement of technology and engineering provides opportunities to explore the idea for the development of an integrated sensing, prediction and alert system of natural disaster events. Therefore, several machine learning techniques have been explored by researchers to harvest knowledge from nature occurrences for prediction of environmental events. The aim of this paper is to investigate, compare and contrast three data modelling techniques, and five data classification training algorithms for spatio-temporal data using real world flood event case study data. A discussion is presented to identify the strengths and weaknesses of each data modelling techniques and training algorithms.

A Framework to Cluster Temporal Data Using Personalised Modelling Approach

This research paper is focused on the framework design of temporal data by using personalised modelling approach in order to cluster the temporal data. Real world problem on flood occurrences is used as a case study focusing only in Malaysia region. The data are designed according to the criteria needed for temporal data clustering, tested with three clustering techniques including K-means, X-means, and K-medoids. Rapid Miner is used for conducting the clustering processes. Finally, the result from each clustering method is compared to conclude and justify the best clustering approach for clustering temporal data.
View Article

An improved computational framework using one stage filtration by incorporating knowledge in gene expression clustering

Abstract- The field of gene expression data analysis has grown in the past few years from being purely data-centric to integrative, aiming at complementing microarray analysis with data and knowledge from diverse available sources. Given a gene expression dataset, the challenge is to predict gene function given the intensity and redundancy of the data and the lack of similarity of expression profiles without degrading the analysis performance and also without assigning genes at random. At the same time, the computational method must be capable of producing highly compacted clusters with furthest separation, high consistency, and accuracy. In this paper, we present a new computational framework for clustering gene expression data. From this experiment, we can conclude that our new framework capable to improve the accuracy thus determined the dominant gene.
View Article

Research on Improving Dominant Yeast Genes for Gene Function Prediction

The field of gene expression data analysis has grown in the past few years from being purely data-centric to integrative, aiming at complementing micro-array analysis with data and knowledge from diverse available sources. Given a gene expression dataset, the challenge is to predict gene function given the intensity and redundancy of the data and the lack of similarity of expression profiles without degrading the analysis performance and also without assigning genes at random. At the same time, the computational method must be capable of producing highly compacted clusters with furthest separation, high consistency, and accuracy. In this paper, we present a new computational framework for clustering gene expression data. From this experiment, we can conclude that our new framework capable to improve the accuracy thus determined the dominant gene.


For a full list of my publications, head to my Google Scholar or visit my ResearchGate profile.

I would also like to introduce you to few of my associates:
Assoc. Professor Dr. Shahreen Kasim - View on Google Scholar
Assoc. Professor Dr. Hairulnizam Mahdin - View on Google Scholar
Dr. Muhaini Othman - View on Google Scholar
Dr. Azizul Azhar Ramli - View on Google Scholar
Professor Nikola Kasabov - View on Google Scholar