Browsing by Author "Yogarajah, B."

Now showing 1 - 2 of 2

Feature selection in automobile price prediction: An integrated approach
(Department of Industrial Management, Faculty of Science, University of Kelaniya Sri Lanka, 2021) Selvaratnam, Sobana; Yogarajah, B.; Jeyamugan, T.; Ratnarajah, Nagulan
Machine learning models for predictions enable researchers to make effective decisions based on historical data. Automobile price prediction studies have been a most interesting research area in machine learning nowadays. The independent variables to model the price and the price predictions are equally important for automobile consumers and manufacturers. Automobile consulting companies determine how prices vary in relation to the independent variables and they can then adjust the automobile's design, commercial strategy, and other factors to fulfill specified price targets. Furthermore, the model will assist management in comprehending a company's pricing patterns. The ability of machine learning systems to predict outcomes is entirely dependent on the effective selection of features. In this paper, we determine the influencing features on automobile price using an integrated approach of LASSO and stepwise selection regression algorithms. We use multiple linear regression to build the model using the selected features. From the experimental results using the automobile dataset from the UCI machine learning repository, the influencing features on automobile price are width, engine size, city mpg, stroke, make, aspiration, number of doors, body style, and drive wheels. Training data accuracy for predicting price was found to be 92%, and testing data accuracy was found to be 87%. The proposed approach supports selecting the most important characteristics of predicting the price of automobiles efficiently and effectively. This research will aid in the development of a model that uses the selected attributes to predict the price of automobiles using machine learning technologies.
Regularization Risk Factors of Suicide in Sri Lanka for Machine Learning
(Faculty of Computing and Technology (FCT), University of Kelaniya, Sri Lanka, 2021) Delpitiya., D. M. A. U.; Kumarage, Prabha M.; Yogarajah, B.; Ratnarajah, Nagulan
Indication to World Health Organization, suicide is a major world public health concern that is in the top twenty leading causes of death worldwide. Sri Lanka is a country that has the highest suicidal rates in the globe. The comprehensive study about risk factors for suicide is important because we can prevent or treat the recognized most risky categories of people. The emergence of big data concepts with machine learning techniques introduced a resurgence in predicting models using risk factors for suicide. Regularization is one of the most decisive components in the statistical machine learning process and this technique is used to reduce the error on the training dataset and prevent over-fitting. Comprehensive regularization approaches are presented here to select significant risk factors for age-specific suicide in Sri Lanka and build unique predictive models. The Least Absolute Shrinkage and Selection Operator (LASSO) approach presents regularization along with the feature selection to improve the prediction precision. The dataset collected for the study is rooted in the Sri Lankan people and the factors used for the analysis are, suicide person’s gender, lived place, education level, mode of suicide, job, reason, suicide time, previous attempts, and marital status. Further, the riskiest age category of the people, who has exposure to suicide, is identified. Multiple linear regression and Ridge regression were used to evaluate the performance of LASSO. The selected most relevant factors with regularization to predict age-specific suicide prove the effectiveness of the proposed regularization approaches.