Statistical and Mathematical Learning: an application to fraud detection and prevention
- Authors: Hamlomo, Sisipho
- Date: 2022-04-06
- Subjects: Credit card fraud , Bootstrap (Statistics) , Support vector machines , Neural networks (Computer science) , Decision trees , Machine learning , Cross-validation , Imbalanced data
- Language: English
- Type: Master's thesis , text
- Identifier: http://hdl.handle.net/10962/233795 , vital:50128
- Description: Credit card fraud is an ever-growing problem. There has been a rapid increase in the rate of fraudulent activities in recent years resulting in a considerable loss to several organizations, companies, and government agencies. Many researchers have focused on detecting fraudulent behaviours early using advanced machine learning techniques. However, credit card fraud detection is not a straightforward task since fraudulent behaviours usually differ for each attempt and the dataset is highly imbalanced, that is, the frequency of non-fraudulent cases outnumbers the frequency of fraudulent cases. In the case of the European credit card dataset, we have a ratio of approximately one fraudulent case to five hundred and seventy-eight non-fraudulent cases. Different methods were implemented to overcome this problem, namely random undersampling, one-sided sampling, SMOTE combined with Tomek links and parameter tuning. Predictive classifiers, namely logistic regression, decision trees, k-nearest neighbour, support vector machine and multilayer perceptrons, are applied to predict if a transaction is fraudulent or non-fraudulent. The model's performance is evaluated based on recall, precision, F1-score, the area under receiver operating characteristics curve, geometric mean and Matthew correlation coefficient. The results showed that the logistic regression classifier performed better than other classifiers except when the dataset was oversampled. , Thesis (MSc) -- Faculty of Science, Statistics, 2022
- Full Text:
- Date Issued: 2022-04-06
- Authors: Hamlomo, Sisipho
- Date: 2022-04-06
- Subjects: Credit card fraud , Bootstrap (Statistics) , Support vector machines , Neural networks (Computer science) , Decision trees , Machine learning , Cross-validation , Imbalanced data
- Language: English
- Type: Master's thesis , text
- Identifier: http://hdl.handle.net/10962/233795 , vital:50128
- Description: Credit card fraud is an ever-growing problem. There has been a rapid increase in the rate of fraudulent activities in recent years resulting in a considerable loss to several organizations, companies, and government agencies. Many researchers have focused on detecting fraudulent behaviours early using advanced machine learning techniques. However, credit card fraud detection is not a straightforward task since fraudulent behaviours usually differ for each attempt and the dataset is highly imbalanced, that is, the frequency of non-fraudulent cases outnumbers the frequency of fraudulent cases. In the case of the European credit card dataset, we have a ratio of approximately one fraudulent case to five hundred and seventy-eight non-fraudulent cases. Different methods were implemented to overcome this problem, namely random undersampling, one-sided sampling, SMOTE combined with Tomek links and parameter tuning. Predictive classifiers, namely logistic regression, decision trees, k-nearest neighbour, support vector machine and multilayer perceptrons, are applied to predict if a transaction is fraudulent or non-fraudulent. The model's performance is evaluated based on recall, precision, F1-score, the area under receiver operating characteristics curve, geometric mean and Matthew correlation coefficient. The results showed that the logistic regression classifier performed better than other classifiers except when the dataset was oversampled. , Thesis (MSc) -- Faculty of Science, Statistics, 2022
- Full Text:
- Date Issued: 2022-04-06
Thicket expansion in a vachellia karroo-dominated landscape and its effect on herbaceous communities
- Authors: Khoza, Marina Rindzani
- Date: 2022-04-06
- Subjects: Savanna ecology South Africa , Forbs South Africa , Grasslands South Africa , Herbaceous plants South Africa , Vegetation dynamics South Africa , Forest canopies South Africa
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/291015 , vital:56808
- Description: Grass and forb species found in savannas are highly diverse, contributing to the structure and function of the savanna system. Where mean annual rainfall is seasonal and high enough to support closed canopy vegetation such as forests or thickets, savannas can exist as an alternative stable state maintained by disturbances such as fire and browsing. Biotic and abiotic processes act on savanna and forest (or thicket) systems maintaining both their tree and herbaceous cover at levels that ensure their persistence in those states. Studies have shown that many semi-arid rangelands in South Africa have undergone a rapid increase in tree cover (of both native and non-native species) over the past several decades. This process of increasing tree cover in semi-arid savannas, termed bush encroachment, results in a biome shift, changing landscapes that were once grasslands with few trees to ones dominated by broad-leaved trees with fewer sun-adapted forbs and grasses. The aim of this study was to investigate the impact of changing woody cover and its associated changes in tree composition, tree canopy structure, light dynamics in the understory and herbaceous community composition on Endwell farm in the Eastern Cape. Canopy cover changes between the years 1949 and 2019 were analysed at 51 sites on the farm and related to historical rainfall patterns. There had been a general increase in tree cover over the past several decades on the farm, and many sites showed a change from open (0-15%) in 1949 to low (1635%), moderate (36-50%) and high (51-100%) canopy cover in 2019. In earlier years most sites had a canopy cover below 50%, and the higher canopy cover values (>65%) occurred in more recent decades. Canopy cover of ~ 50% was found to be rare in each decade. This suggests that ~50% canopy cover maybe a transient, unstable state. The period with the highest rate of canopy cover increase was 2002-2013, and this increase coincided with a high mean annual rainfall 10 years prior to 2002 and a high mean annual rainfall in most years between the 20022013 period. The period between 2002 and 2013 also had the highest number of sites transitioning from lower to higher tree canopy cover classes, indicating that rainfall may have been a factor driving bush encroachment during the past several decades. An increase in canopy cover (a decrease in light transmittance) was accompanied by changes in woody species composition during thicket formation. The low canopy cover (high light transmittance) sites were dominated by Vachellia karroo and Scutia myrtina trees, while high tree cover sites had fewer V. karroo and S. myrtina trees and were rather characterised by an abundance of thicket tree species. Species proportion, NMDS and dendrogram plots indicated that sites with a light transmittance range between 50-100% had similar tree species compositions, different from sites with light transmittances <50%. An increase in tree density was strongly correlated to an increase in canopy cover (from 2019 satellite imagery), density of trees > 3m, maximum height reached by trees, diversity of trees, total canopy volume, total canopy area and leaf area index (LAI), and a decrease in light transmittance. A structural equation model (SEM) was used to explore the relationships between canopy characteristics (maximum canopy area, canopy volume, tree diversity, density of trees, density of trees >3m, individual trees and maximum canopy height), aerial canopy cover in 2019, and light transmittance. The model explained 73% of the variation in light transmittance, mostly via the direct effect of canopy characteristics. Canopy characteristics had a strong influence on both aerial cover in 2019 and directly on light transmittance, but canopy cover in 2019 had a weak influence on light transmittance. The herbaceous layer was rich and dominated by C4 grasses such as Eragrostis plana, Sporobolus fimbriatus, Themeda triandra and Digitaria eriantha) and forbs including Hibiscus aethiopicus, Helichrysum dregeanum, Helichrysum nudifolium and Gerbera viridifolia at low canopy cover sites with high light transmittance. In contrast, high tree cover sites had fewer herbaceous species in general. Grass and forb species characteristic of these sites high canopy cover sites were Panicum maximum, Loudetia flavida, Pellaea viridis and Cyperus spp. Different sites with low light transmittance (<50%) had similar herbaceous species composition. Basal cover, richness, abundance and diversity of herbaceous plants decreased significantly with an increase in tree density, density of trees >3 m, canopy volume, canopy area, canopy cover, LAI, and increased significantly with increasing light transmittance. Most grasses had their highest densities at LAI <0.5, which was estimated to correspond to ~75% light transmittance and ~38% canopy cover and then started to decline thereafter. Herbaceous species basal cover was also highest at LAI <0.5. An SEM model indicated that herbaceous diversity, basal cover and richness responded both to light availability and to the structure of the woody vegetation directly (R2 = 0.53). While the effect of light transmittance on herbaceous communities was strong (0.41), there was little difference between the effect of light transmittance and canopy characteristics (-0.35) on herbaceous communities. Two possible threshold points, relating to two types of transitions in vegetation structure, could be deduced from this study. The first threshold occurred at canopy cover ~ 40% (LAI < ~ 0.5, light transmittance ~ 75%), at which point many of the common herbaceous species, including the dominant C4 grasses, began to decline in abundance while the composition remained characteristic of the savanna state. A canopy cover of less than ~ 40% at a site provides a suitable state for a high abundance of grass and forb species which help maintain an open system by facilitating fires. The second threshold marked a compositional shift between savanna and closed-canopy vegetation states. Savanna species (trees, grasses and forbs) dominated at high light transmittances (>50%) and were significantly reduced at low light transmittances (< 50%), indicating a possible species composition threshold at ~50% light transmittance at which a savanna state switches to a thicket (LAI ~ 1 and canopy cover ~70%). This point indicated the point where there was a significant difference in both tree and herbaceous plant compositions, with a marked reduction in the occurrence of C4 grasses at light transmittance <50%. Fire is supressed when the C4 grass layer is lost, and further thicket encroachment will take place causing complete canopy closure. Land managers in this system should start becoming concerned about a reduction in grass biomass when canopy cover reaches about 40% and would have to reduce tree cover before the threshold of 50% light transmittance (70% canopy cover from aerial photos) is reached to maintain a savanna system. , Thesis (MSc) -- Faculty of Science, Botany, 2022
- Full Text:
- Date Issued: 2022-04-06
Thicket expansion in a vachellia karroo-dominated landscape and its effect on herbaceous communities
- Authors: Khoza, Marina Rindzani
- Date: 2022-04-06
- Subjects: Savanna ecology South Africa , Forbs South Africa , Grasslands South Africa , Herbaceous plants South Africa , Vegetation dynamics South Africa , Forest canopies South Africa
- Language: English
- Type: Academic theses , Master's theses , text
- Identifier: http://hdl.handle.net/10962/291015 , vital:56808
- Description: Grass and forb species found in savannas are highly diverse, contributing to the structure and function of the savanna system. Where mean annual rainfall is seasonal and high enough to support closed canopy vegetation such as forests or thickets, savannas can exist as an alternative stable state maintained by disturbances such as fire and browsing. Biotic and abiotic processes act on savanna and forest (or thicket) systems maintaining both their tree and herbaceous cover at levels that ensure their persistence in those states. Studies have shown that many semi-arid rangelands in South Africa have undergone a rapid increase in tree cover (of both native and non-native species) over the past several decades. This process of increasing tree cover in semi-arid savannas, termed bush encroachment, results in a biome shift, changing landscapes that were once grasslands with few trees to ones dominated by broad-leaved trees with fewer sun-adapted forbs and grasses. The aim of this study was to investigate the impact of changing woody cover and its associated changes in tree composition, tree canopy structure, light dynamics in the understory and herbaceous community composition on Endwell farm in the Eastern Cape. Canopy cover changes between the years 1949 and 2019 were analysed at 51 sites on the farm and related to historical rainfall patterns. There had been a general increase in tree cover over the past several decades on the farm, and many sites showed a change from open (0-15%) in 1949 to low (1635%), moderate (36-50%) and high (51-100%) canopy cover in 2019. In earlier years most sites had a canopy cover below 50%, and the higher canopy cover values (>65%) occurred in more recent decades. Canopy cover of ~ 50% was found to be rare in each decade. This suggests that ~50% canopy cover maybe a transient, unstable state. The period with the highest rate of canopy cover increase was 2002-2013, and this increase coincided with a high mean annual rainfall 10 years prior to 2002 and a high mean annual rainfall in most years between the 20022013 period. The period between 2002 and 2013 also had the highest number of sites transitioning from lower to higher tree canopy cover classes, indicating that rainfall may have been a factor driving bush encroachment during the past several decades. An increase in canopy cover (a decrease in light transmittance) was accompanied by changes in woody species composition during thicket formation. The low canopy cover (high light transmittance) sites were dominated by Vachellia karroo and Scutia myrtina trees, while high tree cover sites had fewer V. karroo and S. myrtina trees and were rather characterised by an abundance of thicket tree species. Species proportion, NMDS and dendrogram plots indicated that sites with a light transmittance range between 50-100% had similar tree species compositions, different from sites with light transmittances <50%. An increase in tree density was strongly correlated to an increase in canopy cover (from 2019 satellite imagery), density of trees > 3m, maximum height reached by trees, diversity of trees, total canopy volume, total canopy area and leaf area index (LAI), and a decrease in light transmittance. A structural equation model (SEM) was used to explore the relationships between canopy characteristics (maximum canopy area, canopy volume, tree diversity, density of trees, density of trees >3m, individual trees and maximum canopy height), aerial canopy cover in 2019, and light transmittance. The model explained 73% of the variation in light transmittance, mostly via the direct effect of canopy characteristics. Canopy characteristics had a strong influence on both aerial cover in 2019 and directly on light transmittance, but canopy cover in 2019 had a weak influence on light transmittance. The herbaceous layer was rich and dominated by C4 grasses such as Eragrostis plana, Sporobolus fimbriatus, Themeda triandra and Digitaria eriantha) and forbs including Hibiscus aethiopicus, Helichrysum dregeanum, Helichrysum nudifolium and Gerbera viridifolia at low canopy cover sites with high light transmittance. In contrast, high tree cover sites had fewer herbaceous species in general. Grass and forb species characteristic of these sites high canopy cover sites were Panicum maximum, Loudetia flavida, Pellaea viridis and Cyperus spp. Different sites with low light transmittance (<50%) had similar herbaceous species composition. Basal cover, richness, abundance and diversity of herbaceous plants decreased significantly with an increase in tree density, density of trees >3 m, canopy volume, canopy area, canopy cover, LAI, and increased significantly with increasing light transmittance. Most grasses had their highest densities at LAI <0.5, which was estimated to correspond to ~75% light transmittance and ~38% canopy cover and then started to decline thereafter. Herbaceous species basal cover was also highest at LAI <0.5. An SEM model indicated that herbaceous diversity, basal cover and richness responded both to light availability and to the structure of the woody vegetation directly (R2 = 0.53). While the effect of light transmittance on herbaceous communities was strong (0.41), there was little difference between the effect of light transmittance and canopy characteristics (-0.35) on herbaceous communities. Two possible threshold points, relating to two types of transitions in vegetation structure, could be deduced from this study. The first threshold occurred at canopy cover ~ 40% (LAI < ~ 0.5, light transmittance ~ 75%), at which point many of the common herbaceous species, including the dominant C4 grasses, began to decline in abundance while the composition remained characteristic of the savanna state. A canopy cover of less than ~ 40% at a site provides a suitable state for a high abundance of grass and forb species which help maintain an open system by facilitating fires. The second threshold marked a compositional shift between savanna and closed-canopy vegetation states. Savanna species (trees, grasses and forbs) dominated at high light transmittances (>50%) and were significantly reduced at low light transmittances (< 50%), indicating a possible species composition threshold at ~50% light transmittance at which a savanna state switches to a thicket (LAI ~ 1 and canopy cover ~70%). This point indicated the point where there was a significant difference in both tree and herbaceous plant compositions, with a marked reduction in the occurrence of C4 grasses at light transmittance <50%. Fire is supressed when the C4 grass layer is lost, and further thicket encroachment will take place causing complete canopy closure. Land managers in this system should start becoming concerned about a reduction in grass biomass when canopy cover reaches about 40% and would have to reduce tree cover before the threshold of 50% light transmittance (70% canopy cover from aerial photos) is reached to maintain a savanna system. , Thesis (MSc) -- Faculty of Science, Botany, 2022
- Full Text:
- Date Issued: 2022-04-06
Protean career management of independent music creatives in Gqeberha : narratives of practice
- Authors: Henneberry, Robyn
- Date: 2021-12
- Subjects: Port Elizabeth (South Africa) , Eastern Cape (South Africa) , South Africa
- Language: English
- Type: Master's theses , text
- Identifier: http://hdl.handle.net/10948/54439 , vital:46541
- Description: This study is concerned with the promotion of local music creatives. The literature explores ways to help the independent music creatives successfully navigate in the music industry by adopting a protean career profile. The research question that was explored in the study was whether or not the independent popular female creatives in the city of Gqeberha’s music industry adopted protean career profiles for successful career navigation. These pop music creatives’ profiles were compared to those of classical creatives for further insight. Relevant literature was analysed to collect data about what international authors have discovered about creatives’ successful navigation of their careers. The information that was gleaned from the analysis indicated that numerous individuals have adopted the profile of the protean career to successfully navigate their environment. The rationale for adopting the protean career profile was that it is directly related to one’s employability. Although the working world (including the music industry) is rapidly changing, this notion of employability will enable an individual to remain updated about the latest trends in their work environment and thus successfully navigate that environment. This concept was applied to research among music creatives in the South African (SA) context. Although there was limited research on the protean career profile, the published research findings that were available indicated that employability and information were in alliance with the international research on the protean career (which constitutes one’s employability). Employing qualitative data collection procedures, the researcher interviewed 14 independent female music creatives with different backgrounds, roles and experiences. Interviews were semi-structured one-on-one as well as focus groups. This narrative strategy of inquiry helped determine whether or not these individuals had adopted the protean career for successful navigation of their working environment. After data collection the researcher engaged in a cyclical coding process for deep analysation of the data utilising atlas.ti. The data was reduced from codes to categories, and eventually towards linking categories. The findings from the study revealed that the participants’ environment (surroundings) did not sufficiently support them, the creatives do not have sufficient guidance in their music paths and not all of them had adopted the protean profile. , Thesis (MA) -- Faculty of Humanities, School of Visual and Performing Arts, 2021
- Full Text:
- Date Issued: 2021-12
- Authors: Henneberry, Robyn
- Date: 2021-12
- Subjects: Port Elizabeth (South Africa) , Eastern Cape (South Africa) , South Africa
- Language: English
- Type: Master's theses , text
- Identifier: http://hdl.handle.net/10948/54439 , vital:46541
- Description: This study is concerned with the promotion of local music creatives. The literature explores ways to help the independent music creatives successfully navigate in the music industry by adopting a protean career profile. The research question that was explored in the study was whether or not the independent popular female creatives in the city of Gqeberha’s music industry adopted protean career profiles for successful career navigation. These pop music creatives’ profiles were compared to those of classical creatives for further insight. Relevant literature was analysed to collect data about what international authors have discovered about creatives’ successful navigation of their careers. The information that was gleaned from the analysis indicated that numerous individuals have adopted the profile of the protean career to successfully navigate their environment. The rationale for adopting the protean career profile was that it is directly related to one’s employability. Although the working world (including the music industry) is rapidly changing, this notion of employability will enable an individual to remain updated about the latest trends in their work environment and thus successfully navigate that environment. This concept was applied to research among music creatives in the South African (SA) context. Although there was limited research on the protean career profile, the published research findings that were available indicated that employability and information were in alliance with the international research on the protean career (which constitutes one’s employability). Employing qualitative data collection procedures, the researcher interviewed 14 independent female music creatives with different backgrounds, roles and experiences. Interviews were semi-structured one-on-one as well as focus groups. This narrative strategy of inquiry helped determine whether or not these individuals had adopted the protean career for successful navigation of their working environment. After data collection the researcher engaged in a cyclical coding process for deep analysation of the data utilising atlas.ti. The data was reduced from codes to categories, and eventually towards linking categories. The findings from the study revealed that the participants’ environment (surroundings) did not sufficiently support them, the creatives do not have sufficient guidance in their music paths and not all of them had adopted the protean profile. , Thesis (MA) -- Faculty of Humanities, School of Visual and Performing Arts, 2021
- Full Text:
- Date Issued: 2021-12
- «
- ‹
- 1
- ›
- »