Philani Khumalo

Dr. Philani Khumalo

orcid

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
23 September 2025 in Journal of Imaging
Verified
Journal Article
Open access
PDF file
TNThandiwe Banethezekile Nyawose
Thandiwe Banethezekile Nyawose
RMRito Clifford Maswanganyi
Rito Clifford Maswanganyi
Philani KhumaloPhilani Khumalo
Philani Khumalo
The early and accurate detection of plant diseases is essential for ensuring food security, enhancing crop yields, and facilitating precision agriculture. Manual methods are labour-intensive and prone to error, especially under varying environmental conditions. Artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), has advanced automated disease identification through image classification. However, challenges persist, including limited generalisability, small and imbalanced datasets, and poor real-world performance. Unlike previous reviews, this paper critically evaluates model performance in both lab and real-time field conditions, emphasising robustness, generalisation, and suitability for edge deployment. It introduces recent architectures such as GreenViT, hybrid ViT–CNN models, and YOLO-based single- and two-stage detectors, comparing their accuracy, inference speed, and hardware efficiency. The review discusses multimodal and self-supervised learning techniques to enhance detection in complex environments, highlighting key limitations, including reliance on handcrafted features, overfitting, and sensitivity to environmental noise. Strengths and weaknesses of models across diverse datasets are analysed with a focus on real-time agricultural applicability. The paper concludes by identifying research gaps and outlining future directions, including the development of lightweight architectures, integration with Deep Convolutional Generative Adversarial Networks (DCGANs), and improved dataset diversity for real-world deployment in precision agriculture.

13 November 2025 in Applied Sciences
Verified
Journal Article
Open access
PDF file
BMBandile Mdluli
Bandile Mdluli
Philani KhumaloPhilani Khumalo
Philani Khumalo
RMRito Clifford Maswanganyi
Rito Clifford Maswanganyi
Brain–Computer Interface (BCI) technology facilitates direct communication between the human brain and external devices by interpreting brain wave patterns associated with specific motor imagery tasks, which are derived from EEG signals. Although BCIs allow applications such as robotic arm control and smart assistive environments, they face major challenges, mainly due to the large variation in EEG characteristics between and within individuals. This variability is caused by low signal-to-noise ratio (SNR) due to both physiological and non-physiological artifacts, which severely affect the detection rate (IDR) in BCIs. Advanced multi-stage signal processing pipelines, including efficient filtering and decomposition techniques, have been developed to address these problems. Additionally, numerous feature engineering techniques have been developed to identify highly discriminative features, mainly to enhance IDRs in BCIs. In this review, several pre-processing techniques, including feature extraction algorithms, are critically evaluated using deep learning techniques. The review comparatively discusses methods such as wavelet-based thresholding and independent component analysis (ICA), including empirical mode decomposition (EMD) and its more sophisticated variants, such as Self-Adaptive Multivariate EMD (SA-MEMD) and Ensemble EMD (EEMD). These methods are examined based on machine learning models using SVM, LDA, and deep learning techniques such as CNNs and PCNNs, highlighting key limitations and findings, including different performance metrics. The paper concludes by outlining future directions.

30 March 2021 in SA Journal of Human Resource Management
Verified
Journal Article
Open access
PDF file
Philani KhumaloPhilani Khumalo
Philani Khumalo
RZRobert W.D. Zondo
Robert W.D. Zondo
Orientation: Global market demands that organisations remain competitive, flexible and relevant in order to adequately react to culturally diverse changes in their human resources. This sentiment describes the importance of cultural diversity in organisations. However, cultural diversity phenomenon has not been given serious attention by business.Research purpose: This study examines the perception on the influence of cultural diversity on business performance in automotive component manufacturers in South Africa.Motivation for the study: Cultural diversity in the organisation has grown over the passage of time with an increase in globalisation. People no longer live in an insular environment. They are part of the worldwide economy competing within a global framework. Hence, cultural diversity becomes a way of creating competitive advantage.Research design, approach and methods: The study was quantitative in design, with middle-level managers in two large automotive component manufacturers in the eThekwini District Municipality in KwaZulu-Natal participating in the study. Of the 168 individuals identified for participation, 118 participated in the study, representing a 70.2% response rate. Descriptive, correlative and t-tests, using the Statistical Package for the Social Sciences (SPSS) were used to analyse the data.Main findings: Study results indicates that a culturally diverse workforce is associated with greater organisational performance, improved strategic decision-making and innovation. Hence, organisations must manage cultural diversity among their workforce for the improvement of business performance.Managerial implications: Culturally diverse workforce are associated with greater organisational performance, improved strategic decision-making and innovation. Hence, management must develop an understanding of their cultural settings if their organisations are to perform effectively.Contribution: The original value of this study is its approach in uncovering the strengths and weaknesses of cultural diversity on business performance in South Africa.

1 December 2018
Verified
Conference Paper
Closed access
Philani KhumaloPhilani Khumalo
Philani Khumalo
BNBakhe Nleya
Bakhe Nleya
AGA. Gomba
A. Gomba
AMAndrew Mutsvangwa
Andrew Mutsvangwa
5G wireless together with optical backbone networks are expected to be the main pillars of the envisaged next /future generation networking (N/FGN) infrastructures. This is an impetus to practical realization of an IoT network that will support and ensure relatively higher bandwidth as well as enhanced quality of service (QoS) in both access and core network sections. The high-speed wireless links at the network peripherals will serve as a conducive platform for device-to-device (D2D) communication. D2D driven applications and services can only be effective as well as secure assuming the associated machine type communication devices (MTCDs) have been successfully verified and authenticated. Typically, D2D type services and applications involve the interaction of several MTCDs in a group. As such, secure and effective D2D group-based authentication and key agreement (AKA) protocols are necessary. They need to inherently achieve efficacy in maintaining the group key unlink-ability as well as generate minimal signalling overheads that otherwise may lead to network congestion. In this paper we detail a secure and efficient Group AKA (Gr-AKA) protocol for D2D communication. Its performance is compared to that of existing similar protocols and is found to comparably lower both computational as well as signalling overhead requirements. Overall the analysis shows that the Gr-AKA protocol improves performance in terms of fulfilling D2D communication's security requirements.

9 November 2019 in Journal of Optical Communications
Verified
Journal Article
Closed access
PDF file
BNBakhe Nleya
Bakhe Nleya
Philani KhumaloPhilani Khumalo
Philani Khumalo
AMAndrew Mutsvangwa
Andrew Mutsvangwa
Heterogeneous IoT-enabled networks generally accommodate both jitter tolerant and intolerant traffic. Optical Burst Switched (OBS) backbone networks handle the resultant volumes of such traffic by transmitting it in huge size chunks called bursts. Because of the lack of or limited buffering capabilities within the core network, burst contentions may frequently occur and thus affect overall supportable quality of service (QoS). Burst contention(s) in the core network is generally characterized by frequent burst losses as well as differential delays especially when traffic levels surge. Burst contention can be resolved in the core network by way of partial buffering using fiber delay lines (FDLs), wavelength conversion using wavelength converters (WCs) or deflection routing. In this paper, we assume that burst contention is resolved by way of deflecting contending bursts to other less congested paths even though this may lead to differential delays incurred by bursts as they traverse the network. This will contribute to undesirable jitter that may ultimately compromise overall QoS. Noting that jitter is mostly caused by deflection routing which itself is a result of poor wavelength and routing assigning, the paper proposes a controlled deflection routing (CDR) and wavelength assignment based scheme that allows the deflection of bursts to alternate paths only after controller buffer preset thresholds are surpassed. In this way, bursts (or burst fragments) intended for a common destination are always most likely to be routed on the same or least cost path end-to-end. We describe the scheme as well as compare its performance to other existing approaches. Overall, both analytical and simulation results show that the proposed scheme does lower both congestion (on deflection routes) as well as jitter, thus also improving throughput as well as avoiding congestion on deflection paths.

Verified
Journal Article
Open access
PDF file
KRKeenan Ryan Ramnarain
Keenan Ryan Ramnarain
RMRito Clifford Maswanganyi
Rito Clifford Maswanganyi
Philani KhumaloPhilani Khumalo
Philani Khumalo
Alzheimer’s disease (AD) is the most prevalent neurodegenerative disorder worldwide, affecting approximately 56.9 million people in 2021 and projected to reach 152 million by 2050. Its defining pathological features, amyloid-beta plaques and neurofibrillary tangles, accumulate for up to two decades before cognitive symptoms emerge, placing the preclinical and mild cognitive impairment (MCI) stages at the centre of the early detection problem. Despite this, current diagnostic practice in routine clinical settings remains unreliable, with post-mortem studies placing the specificity of clinical AD diagnosis between 44.3 and 70.8% even in specialist memory clinics. Machine learning (ML) and deep learning (DL) applied to neuroimaging and electrophysiological data have emerged as candidate tools for closing this diagnostic gap, yet whether the accuracy figures reported in published studies translate into clinically useful performance on independent data remains unresolved. This study presents a structured comparative review of machine learning and deep learning methods reported across four publicly available Alzheimer’s disease datasets, namely the Alzheimer’s Disease Neuroimaging Initiative (ADNI), the Open Access Series of Imaging Studies (OASIS), the OpenNeuro ds004504 electroencephalography (EEG) dataset, and the Kaggle Alzheimer’s magnetic resonance imaging (MRI) dataset. Thirteen model families are examined through the published literature rather than through new experiments, and for each model and dataset combination, the best accuracy reported in the source study is recorded alongside the model’s mathematical formulation. All performance figures reported in this abstract and throughout the paper are taken from the published studies reviewed, not from new experiments conducted by the authors. Across the reviewed studies, deep learning architectures pre-trained on ImageNet and fine-tuned on neuroimaging data are reported to produce the highest accuracy on MRI classification tasks. Residual Network (ResNet)-101 is reported at 98.21 percent on ADNI and 97.45 percent on OASIS, while the IncepRes fusion architecture reaches 98.35% on OASIS by combining multi-scale feature extraction from InceptionV3 with residual connectivity from ResNet152V2. Traditional machine learning classifiers remain competitive on tabular clinical and biomarker data, with Extreme Gradient Boosting (XGBoost) reaching 91% on ADNI multiclass features. Logistic Regression achieves 82 to 85% on binary ADNI classification and is the only classifier in this review that provides explicit per-feature prediction contributions without post hoc tooling. Gaussian Naïve Bayes achieves 80 to 83% on the same task. On the OpenNeuro EEG dataset, K-nearest neighbours (KNN) with singular value decomposition (SVD) entropy features achieves 91% binary accuracy, with feature engineering quality determining performance more reliably than classifier architecture. Eight principal findings emerge from the cross-dataset analysis. Binary classification consistently outperforms multiclass by 10 to 30% across all datasets, reflecting the genuine biological ambiguity of the mild cognitive impairment category. Dataset size and augmentation predict reported accuracy more reliably than model architecture. Ensemble methods outperform individual classifiers by 5 to 8% in nearly every imaging study. Deeper architectures can overfit small clinical cohorts. EEG models trail MRI models by approximately 10 to 15% on comparable binary tasks. Cross-dataset generalisation has not been systematically evaluated in most studies, and the few that have tested it report accuracy drops of 5 to 10% or more when models encounter data from different scanners or cohorts. Eight recurring limitations constrain the clinical utility of these findings. Small sample sizes and limited demographic diversity, severe class imbalance inflating raw accuracy metrics, poor cross-dataset generalisation driven by scanner heterogeneity, limited deep learning interpretability, the dominance of binary over multiclass tasks, the absence of longitudinal modelling despite available datasets, inadequate standardisation of preprocessing and evaluation protocols, and the signal-to-noise ratio constraints specific to EEG recordings of elderly patients collectively define the gap between benchmark performance and clinical readiness. Future work must prioritise multi-centre training cohorts, multimodal fusion architectures, longitudinal progression modelling, and standardised interpretability evaluation as non-optional requirements for any system intended for clinical deployment.