https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4577 2026-04-24T11:24:59Z 2026-04-24T11:24:59Z Wasala, W M V D https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4868 2025-07-08T05:20:05Z 2024-09-18T00:00:00Z

Title: Future of Sri Lankan Apparel Industry: Proposal for the B2B Sales Trend Analysis Using Machine Learning Approach Authors: Wasala, W M V D Abstract: ABSTRACT In the backdrop of global economic challenges, Sri Lanka's apparel export industry, a significant contributor to the nation's economy, faces threats amidst the country's severe economic crisis. Despite its reputation for ethical sourcing and high-quality garments, Sri Lanka's market share in the global garment industry is relatively small compared to the dominating country, China. Recognizing the challenges associated with this reduced market share, Expo Group of Industries, a leading engineering plant in Sri Lanka, acknowledges the necessity of adopting a data-driven approach to navigate complexities and maintain competitiveness. The company is committed to leveraging data-driven strategies to overcome industry challenges, ensuring it can continue to provide tailored solutions for its clients in the fashion industry. Sri Lanka is grappling with a severe economic crisis since March 2022, marked by a drastic drop in foreign reserves and a subsequent impact on industries, notably the apparel sector. The crisis, rooted in a dollar shortage and exacerbated by electricity tariff hikes and unfavorable tax policies, has led to increased production costs, shipping challenges, and delays in order fulfillment. The political and economic instability has eroded trust among foreign buyers, resulting in reduced orders and job losses in the apparel industry. Amid these challenges, a proposed research project aims to develop a tailored forecasting model using machine learning and time series analysis to improve B2B sales predictions in the Sri Lankan apparel industry, addressing a critical knowledge gap and offering practical insights for industry stakeholders. The study investigates sales forecasting techniques in the B2B apparel industry, revealing that SARIMAX, Random Forest Regression, and XGBoost are effective models. While LSTM lags due to data limitations, Random Forest Regression and XGBoost consistently outperform ARIMA-based models, with XGBoost emerging as the superior performer based on lower MSE, higher R2, and Explained Variance Score. These findings align with prior research highlighting the efficacy of machine learning models in sales prediction. The study fills gaps in B2B sales forecasting literature for the apparel industry, emphasizing the importance of data-driven decision-making and customer profiling for maximizing financial performance. Despite limitations, the research provides a robust foundation for evidence-based decision-making in navigating challenges and capitalizing on opportunities in the Sri Lankan apparel industry.

2024-09-18T00:00:00Z Rajapakse, R.W.U.S https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4867 2025-07-08T05:18:06Z 2024-09-24T00:00:00Z

Title: A Model to Predict Daily Gold Price Using Machine Learning-Based Predictive Analysis Approach Authors: Rajapakse, R.W.U.S Abstract: Abstract This thesis presents a comprehensive analysis of daily gold price prediction in the Sri Lankan market, utilizing a range of exogenous variables to enhance forecasting accuracy. The study explores the impact of key economic indicators, including Brent crude oil prices, USD to LKR exchange rates, CNY to LKR exchange rates, silver prices, S&P 20 index, Colombo Consumer Price Index (CCPI), and gold reserves on the volatility and trends of gold prices in Sri Lanka. Two powerful machine learning algorithms, XGBoost and Random Forest, were employed to predict daily gold prices. The study investigates the predictive performance of these algorithms and compares their effectiveness in capturing the intricate dynamics of the Sri Lankan gold market. The models were trained and tested on a dataset spanning from January 2014 to September 2022 to ensure robustness and reliability in the results. The findings reveal that XGBoost outperformed Random Forest in terms of predictive accuracy and model performance. The superiority of XGBoost suggests its efficacy in handling complex relationships and nonlinear patterns within the dataset, thereby providing more accurate and reliable predictions of daily gold prices in the Sri Lankan market. Furthermore, the inclusion of diverse exogenous variables allows for a more holistic understanding of the factors influencing gold prices. The study contributes valuable insights to the financial community, policymakers, and investors. Keywords: Gold Price, XGBoost, Radom Forest, Machine Learning

2024-09-24T00:00:00Z Fazla, N. F. https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4866 2025-07-08T05:15:50Z 2024-09-29T00:00:00Z

Title: Prediction of ICU Readmissions using LSTM in Low and Middle Income Countries Authors: Fazla, N. F. Abstract: ABSTRACT This study addresses the persistent challenge of Intensive Care Unit (ICU) readmission, focusing on the unique context of Lower and Middle Income Countries (LMICs). Despite advancements in medical technology, ICU readmissions remain a critical issue, with implications for healthcare resources and patient outcomes. To address the challenge of ICU readmission, accurate prediction models are needed to identify patients at high risk of readmission because the prediction of readmission before the patient is discharged, will help physicians re-evaluate the discharge of the patient and reduce the immature discharges. The existing literature predominantly stems from high-income countries (HICs), and this study aims to fill the gap by developing a predictive model tailored to LMICs context. It utilizes the Long Short-Term Memory (LSTM), known for its ability to capture temporal dependencies in sequential patients’ data to predict the early ICU readmission (readmission within 48 hours followed by index discharge) of the patients and feature ablation test to extract the important factors associated with ICU readmission. 2.85% (306) of discharges to the wards were later readmitted within 48 hours to the intensive care unit. The LSTM model with a cost-sensitive training had significantly better performance (area under the receiver operating curve, 0.68) compared to the baseline models with traditional machine learning approaches. It highlights that the deep learning models improve the accuracy of decision-making in predicting ICU readmission.

2024-09-29T00:00:00Z KAPILABANDARA, W.M.U.C. https://dl.ucsc.cmb.ac.lk/jspui/handle/123456789/4865 2025-07-08T05:12:56Z 2024-10-16T00:00:00Z

Title: Leveraging Data Analytics for Lapse Reduction in Life Insurance Authors: KAPILABANDARA, W.M.U.C. Abstract: Abstract Insurance companies operate on the concept of pooling losses among their insureds. An insurer invests the premiums to earn enough money not only to pay for losses but also to operate the company and gain a profit. Thus, the insurance company must reasonably predict the payments that will be made for loss and charge affordable premiums to insure a risk. The term "lapse" refers to the termination of an insurance policy by the policyholder for any reason other than the death of the policyholder. When an insurance policy lapses, it will decrease the performance of the product, the initial year’s expense of the policy may not be covered and it will create a loss of public image. Since retaining existing customers is much cheaper and more profitable than getting a new customer, it is crucial to identify policies which are likely to lapse. Even though the insurance industry uses a number of mathematical, statistical, and financial concepts to understand the behaviour of policyholders and quantify future liabilities and risks, those approaches have major drawbacks. This study focuses on predicting individual policyholder lapse rate and identifying scenarios which reduce lapses in Sri Lankan Insurance industry. To conduct this, firstly data of policyholder need to be collected and analysed. This information is gathered form an Insurance Company. Policies that commence from 2013 to 2022 are included. 32 parameters are considered for the analysis and variable importance is calculated. Then Random Survival Forest (RSF) and Cox net Survival Analysis are used to predict the lapse rate. Those techniques let the model to construct survival functions with different shapes for each insured. Model performance is high in random survival forest compared to cox net survival analysis as it captures linear, non-linear relationships and interactions between many factors. Hence variable importance is calculated using random survival forest. Then scenarios are performed to identify policy characteristics which give low lapse rate, in other words high survival rate. By the findings, it was successfully concluded that, through machine learning teachings, insurance companies will not only be able to predict the lapse rate of individual policyholders but also be able to identify policy characteristics that give a high survival rate.

2024-10-16T00:00:00Z