Data Analytics and Applied Mathematics (DAAM)

Customer analytics for online retailers using weighted k-means and RFM analysis

A. Serwah, K.W. Khaw, S.P.Y. Cheang, A. Alnoor — Sun, 30 Apr 2023 00:00:00 +0000

In recent years, there has been a significant trend toward data-driven enterprises in the business world. This trend is exemplified by the frustration reported by 74% of customers when they encounter ads that are not relevant to them, as reported by Infosys. This emphasizes the importance of personalization in marketing efforts. In order to effectively personalize marketing efforts, businesses often track and analyze the actions of consumers when they interact with websites or click on ads. However, creating completely personalized content for every individual is not practical due to the vast number of people and limited resources and time. In this study, a new approach has been used to segment customers based on the combination of RFM analysis and weighted k-means clustering to help an online retailer better target its customers. The results with weighted k-means are significantly higher with a silhouette score of 0.40 compared to 0.30 of the traditional k-means.

Prediction of US airline passenger satisfaction using machine learning algorithms

A.C.Y. Hong, K.W. Khaw, X.Y. Chew, W.C. Yeong — Sun, 30 Apr 2023 00:00:00 +0000

Due to the COVID-19 pandemic, the U.S. financial system and economy have also been severely affected. The U.S. airline industry has been hit particularly hard by the COVID-19 pandemic. Additionally, the aviation industry is also full of competition. One of the ways to attract customers and compete with other airline companies is by improving their service quality. Therefore, this study aims to predict the satisfaction of airlines based on the machine learning model and discover which features are more correlated with the target variable. In this study, the dataset consists of 129,880 observations and 1 target, 22 features or attributes (not including the identification). In this study, the result showed that the features that slightly correlate more with customer satisfaction are 'Online boarding', 'Inflight entertainment', 'Seat comfort', 'On-board service', 'Leg room service', 'Cleanliness', 'Flight Distance' and 'Inflight wifi service'. Then, K-Nearest Neighbors (KNN), Decision Tree Classifier (DTC), Logistic Regression (LR), Random Forest (RF), Naïve Bayes (NB) and AdaBoost were used to build the classification models. Data cleaning, exploratory data analysis, feature selection and One Hot Encoding were also performed before building the models. Finally, the models were evaluated based on their accuracy, precision, recall and F1-score. The results suggest that the champion model for this study is Random Forest, which achieved 89.20% accuracy, 93.04% precision and 88.80% F1-score. The results of this study can be used as a guide in applying machine learning to predict the satisfaction of airline passengers. This can also contribute to attracting passengers by improving the airline service quality.

The implementation of conjugate gradient methods for data fitting

Sun, 30 Apr 2023 00:00:00 +0000

The conjugate gradient (CG) method is widely used to solve the unconstrained optimization problem by finding the optimal solution. This problem can be solved by an iterative method. CG method can be classified into classical, modified, spectral, three terms, and hybrid. In this research, Polak-Ribiere-Polyak (PRP), Rivaie-Mustafa-Ismail-Leong (RMIL), Nurul Hajar-Mustafa-Rivaie (NMR) and Linda-Aini-Mustafa-Rivaie (LAMR) are the four chosen methods for this comparison study. These methods are tested under the Armijo line search. There are 14 test functions with five initial points and various variables are chosen. This comparison study is tested using MatlabR2022a to evaluate iteration number and CPU time. The performance profiles of the numerical result are plotted using a Sigma plot. Then, a set of data, the ASB dividend rate is used to form a linear model. In conclusion, PRP performs better than any other method since it yields the best numerical results and is applicable for data fitting.

Comparison of recurrent neural network and long-short term memory technique in predicting mortality rate in Malaysia

S. Anupriya, N.A. Ramli, L.J. Awalin — Sun, 30 Apr 2023 00:00:00 +0000

The Disease Control Division, Ministry of Health states that mortality data are a crucial tool for examining a community's health. Several studies have been conducted on developing a predictive model to predict the mortality rate but most of them are applying the standard machine learning models. Hence, this study is conducted with its main objective to develop the Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) models in predicting the general mortality rate in Malaysia and compare the predicted mortality rates between data with and without post-pandemic statistics. The death rate was investigated using the 2022 Revision of World Population Prospects, the twenty-seventh edition of official United Nations population estimates and projections issued by the Population Division of the United Nations Secretariat's Department of Economic and Social Affairs. A Z-test was employed to investigate if the 2021 and 2022 mortality rates bring any significant difference to the predicted mortality rates. The findings confirm the superiority of the LSTM model in accurately predicting mortality rates compared to RNN with the inclusion of post-pandemic data did not significantly affect the model's predictions. Overall, this study contributes to the understanding of Malaysian mortality rates and provides a foundation for future investigations in this field. The accuracy in predicting the future mortality rates, particularly in pandemic scenarios, remains challenging and requires further research.

The development of a predictive model for students’ final grades using machine learning techniques

N.H.A Rahman, S.A. Sulaiman, N.A. Ramli — Sun, 30 Apr 2023 00:00:00 +0000

ABSTRACT – As per research, utilizing predictive analytics in education can be very beneficial. It can help educators improve students' performance by analyzing historical data through various approaches such as data mining and machine learning. However, there is a scarcity of studies on using machine learning and predictive analytics to enhance student performance in Malaysian higher education. This study used the records of 450 students enrolled in the Business Statistics course at Universiti Islam Pahang Sultan Ahmad Shah (UnIPSAS) from 2013, obtained from UnIPSAS's Learning Management System. The aim was to develop the best predictive model for forecasting students' final grades based on their performance levels, using machine learning techniques such as Decision Tree, k-Nearest Neighbor, and Naïve Bayes. The final model was developed using Python software. The results showed a strong negative correlation between the students' carry marks and their final grades, with an r-value of -0.8. Naïve Bayes was found to be the best model, with an AUC score of 0.79.

Simulation and visualization of wave equation

N.A. Salahudin, N. Roslee, N. Zull Pakkal, S. Zokree, H.F. Saipan@Saipol — Sun, 30 Apr 2023 00:00:00 +0000

The partial differential equation (PDE) is significant in mathematics, physics, and engineering fields. The PDE model is one of the problems in a real-life situation that is complex to resolve and it involves many variables to be solved. The wave equation is one of the second-order partial differential equations that is commonly used in mathematics and physics. This equation was first studied in the 18^th century by Euler, d'Alembert, Lagrange, and Laplace. In this paper, the finite difference method is used to solve the simple equation and compared with theoretical calculation. The results for the function are verified using Visual C++ programming. The values of error from both methods are compared, which are very small. The results show that the numerical calculation is as accurate as the theoretical method.