Browsing by Author "Dissanayake, D. M. L. M."

Now showing 1 - 2 of 2

An expert system for legal counselling in Sri Lanka
(Research Symposium on Pure and Applied Sciences, 2018 Faculty of Science, University of Kelaniya, Sri Lanka, 2018) Dissanayake, D. M. L. M.; Rajapakse, R. A. C. P.
According to the Human Right Commission of Sri Lanka (HRCSL) publications and their reports, lack of knowledge is the severe problem in Sri Lankan legal system. In addition to this the major reason, lack of resources, lack of awareness raising program on legal matters, lack of confidence of public on presenting their legal problems are some minor reasons. Due to these problems, most of the Sri Lankans face many legal troubles in their day today life. They do not have clear understanding about their legal matters. To reduce these legal problems, they need to have a proper and efficient legal counseling service of an expert legal officer or an expert lawyer. Legal Aid Commission (LAC) is the main legal counselling provider in Sri Lanka. LAC has expert Legal officers to provide these services. However, the commission has limited number of resources. Every counsellor has a huge work load. In addition to the LAC, Institution of human rights and National Child Protection Authority also provide legal counselling. Although there are many legal counselling institutions in Sri Lanka, a decrease of legal problems cannot be observed. Due to these problems, there is an urgent need to develop a legal expert system for the people to get advised for their legal troubles. An expert system is a computer system which is having an ability to reproduction of logical decision-making process of human experts. It consists of a knowledge base and an inference engine. The goal of this study is to develop a software system to assist the preliminary counselling process. It will provide a primary solution to start their solving process of legal problems. This context considers only about the children’s and the women’s rights violations. Extracting the knowledge of the experts in relevant field by observing their counselling process is the first step for achieving this target. Main purpose is to identify the conditions that experts mainly consider when solving legal problems. On the knowledge of legal counselling experts, a rule-based engine has been implemented. For this purpose, information about the selected scenarios are represented by a series of if-then statements. Developing an expert system with a web interface is then carried out. Finally, it is to get the expert users’ feedback to arrive conclusion. Using this system end users can get guidance to their legal problem by selecting one or more answers from a list or by entering data. The developed software system will contribute to increase the legal awareness of the Sri Lankan people by providing a primary solution to their legal matters.
Machine learning model to predict bank customer's next expenditure with relevant merchant category
(Faculty of Science, University of Kelaniya Sri Lanka, 2023) Umayanga, A. M. K. H.; Dissanayake, D. M. L. M.
The banking industry's increasing reliance on debit card transactions has generated a wealth of valuable data for understanding consumer behaviour. This study aims to develop a machine learning model to predict a customer's next expenditure and the corresponding merchant category using 50 customers' debit card transaction data for 11 years. Unlike existing research focused on bankrupt users and fraud detection, this study addresses the next expenditure prediction with merchant categories. For the bank, predicting a customer's next expenditure and merchant category enables targeted marketing efforts. The bank can send alert messages with discount offers specifically to each customer's spending habits, reducing marketing costs by only targeting relevant customers for relevant merchant types. Additionally, customers benefit from early reminders, allowing them to manage their finances effectively. For instance, a customer can receive a reminder about an upcoming insurance payment and allocate funds, accordingly, avoiding unnecessary expenses. This proactive approach can help reduce the number of bankrupt customers and long-term customer relationships. Challenges in this study include obtaining a dataset that is not readily available on the internet. The dataset was provided by the Digital Banking Department at the Head Office of the People's Bank while ensuring data privacy. Data preprocessing involved removing null values and unnecessary columns and creating customer IDs instead of account numbers. Then, identified 36 customers who consistently used debit cards and categorised merchant names into 11 groups. The dataset was split into training and testing sets using a specific date. Three machine learning algorithms, gradient boosting regressor, random forest regressor, and random forest classifier, were employed. Gradient boosting regressor is used to predict expenditures and merchant categories after encoding the categories using one-hot encoding. Random forest regressor is for expenditure prediction, and random forest classifier is used for merchant category prediction. Ordinal encoding was used to convert categories into numerical values. Model performance was optimised through hyperparameter (learning rate, number of trees, maximum depth of each decision tree, minimum number of samples required to split an internal node, minimum number of samples required to be at a leaf node, and fixed random seed for reproducibility) tuning using grid search, evaluating various combinations of hyperparameters through cross-validation. Models run through each customer’s unique dataset since expanding patterns are different from each other. The results showed that the random forest regressor and random forest classifier-based method achieved higher accuracy compared to the gradient boosting regressor. This was evident from R2 scores (0.9866 and 1.0605) and mean squared error values (MSEs are 313165.9622 and 5.6257). However, the method yielded R2 scores exceeding 1 and a high MSE value due to an unbalanced dataset, where customers' debit card usage frequency varied. Obtaining a balanced dataset with an equal number of transactions for each customer is challenging, especially when requesting data from a bank. In the future, this study could be extended to predict the exact time and date of transactions using techniques like long short-term memory (LSTM) with a larger dataset like 1000 customers.