Faculty of Accounting and Informatics
Permanent URI for this communityhttp://ir-dev.dut.ac.za/handle/10321/1
Browse
2 results
Search Results
Item Data mining and machine learning : a study of the CO2 emission trends in South Africa(2024) Mohamed, Ghulam Masudh; Patel, Sulaiman Saleem; Naicker, NalindrenThis study addresses the pressing global issue of elevated carbon dioxide emissions (CO2E), with a particular focus on South Africa (SA), which ranks amongst the world's top emitters and largest in Africa. By introducing a novel integration of Change-point Analysis (CPA) and Machine Learning (ML) techniques, this research addresses significant gaps in CO2E trend analysis. Unlike previous studies, this research applies CPA methodologies within the distinct context of SA, employing algorithms like cumulative sum (CUSUM) and Bootstrap analysis to pinpoint crucial change-points in CO2E data specific to the country. The Bootstrap analysis determines the confidence levels associated with each detected change. Additionally, this study sought to validate historical trends and predict future patterns using ML models, with a specific focus on employing the AdaBoost ensemble learning technique. Drawing on insights from a Preferred Reporting Items for Systematic Reviews and MetaAnalyses (PRISMA)-based systematic review, the research selects input variables based on the factors identified as significant contributors to CO2E, ensuring the models capture the relevant variables effectively. The results of the systematic review highlight energy production and economic growth as key drivers of CO2E, thus validating their selection as input data for constructing the CPA and ML models. To conduct this study, secondary data was obtained from the World Bank's Open Data initiative data repository, a common source for environmental research. This selection was justified by a literature review, which highlighted the reliability and applicability of this data source. The CPA results reveal significant change-points in electricity generation, economic growth, and CO2E, with an average confidence level of 94%, indicating the accuracy of this analytical approach. Moreover, the CPA results emphasise the relationship between economic growth, electricity production, and CO2E in SA. Before forecasting future CO2E trends, the effectiveness of the AdaBoost regressor in enhancing model performance was benchmarked against traditional ML algorithms, including Linear regression, Polynomial regression, Bayesian Linear regression and K-Nearest Neighbors (KNN) regression, to determine the most effective technique for forecasting CO2E. The researcher evaluated model performance using key regression ML performance metrics, including Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), coefficient of determination (R2) score, and an additional accuracy score introduced by the researcher. Notably, the AdaBoost models demonstrated superior performance, with an average RMSE score of 10,143.17 kilotons (kt), MAE score of 9,642.64 kt, R2 of 0.90, and accuracy of 96.74%. The study also revealed that, on average, models that were trained using the AdaBoost algorithm surpassed traditional ML models, in terms of performance. They achieved a reduction in RMSE score by 6,417.29 kt, a decrease in MAE score by 4,358.09 kt, an increase in R2 score by 0.07 and enhanced accuracy by 0.60%. Additionally, a comparative analysis of the repeated holdout methods and cross-validation techniques was conducted, with results revealing that repeated holdout had a more significant impact on model performance. After excluding outliers, the average improvement in crossvalidation results, due to the repeated holdout method, was a decrease of 783.32 kt for RMSE, a reduction of 1,289.39 kt for MAE, and an increase of 0.88% for accuracy. The extent to which the repeated holdout method improved the performance of ML models that were integrated with cross-validation techniques, was correlated with the initial model performance. For ML models with RMSE and MAE scores equal to or exceeding 15,000 kt, the findings indicate that the repeated holdout methods studied should enhance performance by at least 2,000 kt. Similarly, an improvement of nearly 3% or higher in accuracy was noted, when the crossvalidation value for this metric was 94% or lower. The AdaBoost model, integrated with repeated holdout, was selected as the optimal model, as evidenced by the results, for forecasting CO2E in SA from 2021 to 2027. The forecasted CO2E trends validate that energy production and economic growth are indeed the primary drivers of CO2E in SA, as previously highlighted by the CPA model. This underscores the importance of addressing these factors to effectively mitigate carbon emissions in the country. Moreover, the forecasted results indicate that SA is unlikely to meet the global temperature limit of 1.5 degrees Celsius by 2030, given the trajectory showing a shortfall in achieving the target level of 334 million tonnes (Mt) of CO2E, agreed upon in the Paris Agreement. However, the country did meet its CO2E commitments outlined in the 2030 National Development Plan, showing some progress towards environmental sustainability. Nonetheless, the failure to meet these targets at their lower ranges suggests the need for further efforts to reduce carbon emissions, which is crucial for aligning with the Paris Agreement objectives and achieving a zero net emission rate by 2050. This highlights the importance of ongoing initiatives to enhance environmental policies and practices in SA. Future research should focus on integrating load-shedding dynamics into the analysis to examine and confirm its effects on energy production, economic growth, and CO2E in SA. Additionally, future research should focus on forecasting future change-points for the socio-economic indicators or variables utilised in this study. This can help policymakers anticipate fluctuations and devise proactive strategies, to address environmental and economic challenges effectively. It is also recommended that future research consider the output of renewable energy production, when analysing CO2E trends.Item Data augmentation for deep learning algorithms that perform driver drowsiness detection(The Science and Information Organization, 2023-01) Mohamed, Ghulam Masudh; Patel, Sulaiman Saleem; Naicker, NalindrenDriver drowsiness is one of the main causes of driver-related motor vehicle collisions, as this impairs a person’s concentration whilst driving. With the enhancements of computer vision and deep learning (DL), driver drowsiness detection systems have been developed previously, in an attempt to improve road safety. These systems experienced performance degradation under real-world testing due to factors such as driver movement and poor lighting. This study proposed to improve the training of DL models for driver drowsiness detection by applying data augmentation (DA) techniques that model these real-world scenarios. This paper studies six DL models for driver drowsiness detection: four configurations of a Convolutional Neural Network (CNN), two custom configurations as well as the architectures designed by the Visual Geometry Group (VGG) (i.e. VGG16 and VGG19); a Generative Adversarial Network (GAN) and a Multi-Layer Perceptron (MLP). These DL models were trained using two datasets of eye images, where the state of eye (open or closed) is used in determining driver drowsiness. The performance of the DL models was measured with respect to accuracy, F1-Score, precision, negative class precision, recall and specificity. When comparing the performance of DL models trained on datasets with and without DA in aggregation, it was found that all metrics were improved. After removing outliers from the results, it was found that the average improvement in both accuracy and F1 score due to DA was +4.3%. Furthermore, it is shown that the extent to which the DA techniques improve DL model performance is correlated with the inherent model performance. For DL models with accuracy and F1-Score ≤ 90%, results show that the DA techniques studied should improve performance by at least +5%