COVID-19 South African Vaccine Hesitancy Models Show Boost in Performance Upon Fine-Tuning on M-pox Tweets (preprint)

Very large numbers of M-pox cases have, since the start of May 2022, been reported in non-endemic countries leading many to fear that the M-pox Outbreak would rapidly transition into another pandemic, while the COVID-19 pandemic ravages on. Given the similarities of M-pox with COVID-19, we chose to test the performance of COVID-19 models trained on South African twitter data on a hand-labelled M-pox dataset before and after fine-tuning. More than 20k M-pox-related tweets from South Africa were hand-labelled as being either positive, negative or neutral. After fine-tuning these COVID-19 models on the M-pox dataset, the F1-scores increased by more than 8% falling just short of 70%, but still outperforming state-of-the-art models and well-known classification algorithms. An LDA-based topic modelling procedure was used to compare the miss-classified M-pox tweets of the original COVID-19 RoBERTa model with its fine-tuned version, and from this analysis, we were able to draw conclusions on how to build more sophisticated models.

Detecting the Presence of COVID-19 Vaccination Hesitancy from South African Twitter Data Using Machine Learning (preprint)

Very few social media studies have been done on South African user-generated content during the COVID-19 pandemic and even fewer using hand-labelling over automated methods. Vaccination is a major tool in the fight against the pandemic, but vaccine hesitancy jeopardizes any public health effort. In this study, sentiment analysis on South African tweets related to vaccine hesitancy was performed, with the aim of training AI-mediated classification models and assessing their reliability in categorizing UGC. A dataset of 30000 tweets from South Africa were extracted and hand-labelled into one of three sentiment classes: positive, negative, neutral. The machine learning models used were LSTM, bi-LSTM, SVM, BERT-base-cased and the RoBERTa-base models, whereby their hyperparameters were carefully chosen and tuned using the WandB platform. We used two different approaches when we pre-processed our data for comparison: one was semantics-based, while the other was corpus-based. The pre-processing of the tweets in our dataset was performed using both methods, respectively. All models were found to have low F1-scores within a range of 45$\%$-55$\%$, except for BERT and RoBERTa which both achieved significantly better measures with overall F1-scores of 60$\%$ and 61$\%$, respectively. Topic modelling using an LDA was performed on the miss-classified tweets of the RoBERTa model to gain insight on how to further improve model accuracy.

Development of an Early Alert System for an Additional Wave of COVID-19 Cases using a Recurrent Neural Network with Long Short-Term Memory (preprint)

The impact of the still ongoing “Coronavirus Disease 2019” (COVID-19) pandemic has been and is still vast, affecting not only global human health and stretching healthcare facilities, but also profoundly disrupting societal and economic systems worldwide. The nature of the way the virus spreads causes cases to come in further recurring waves. This is due a complex array of biological, societal and environmental factors, including the novel nature of the emerging pathogen. Other parameters explaining the epidemic trend consisting of recurring waves are logistic-organizational challenges in the implementation of the vaccine roll-out, scarcity of doses and human resources, seasonality, meteorological drivers, and community heterogeneity, as well as cycles of strengthening and easing/lifting of the mitigation interventions. Therefore, it is crucial to be able to have an early alert system to identify when another wave of cases is about to occur. The availability of a variety of newly developed indicators allows for the exploration of multi-feature prediction models for case data. Ten indicators were selected as features for our prediction model. The model chosen is a Recurrent Neural Network with Long Short-Term Memory. This paper documents the development of an early alert/detection system that functions by predicting future daily confirmed cases based on a series of features that include mobility and stringency indices, and epidemiological parameters. The model is trained on the intermittent period in between the first and the second wave, in all of the South African provinces.

Big Data- and Artificial Intelligence-Based Hot-Spot Analysis of COVID-19: Gauteng, South Africa, as a case study (preprint)

“Coronavirus Disease 2019” (COVID-19) related data contain many complexities that must be taken into account when extracting information to guide public health decision- and policy-makers. In generalising the spread of a virus over a large area, such as a province, it must be assumed that the transmission occurs as a stochastic process. This statistically random spread of a virus through a population is the core of the majority of Susceptible-Infectious-Recovered-Deceased (SIRD) models and is dependent on factors such as number of infected cases, infection rate, level of social interactions, susceptible population and total population. However, the spread of COVID-19 and, therefore, the data representing the virus progression do not always conform to a stochastic model. In this paper, we have focused on the most influential non-stochastic dynamics of COVID-19, hot-spots, utilizing artificial intelligence (AI) based geo-localization and clustering analyses, taking Gauteng (South Africa) as a case study.

Leveraging Artificial Intelligence and Big Data to Optimize COVID-19 Clinical Public Health and Vaccination Roll-Out Strategies in Africa (preprint)

COVID-19 is imposing massive health, social and economic costs. While many developed countries have started vaccinating, most African nations are waiting for vaccine stocks to be allocated, and are using clinical public health (CPH) strategies to control the pandemic. The emergence of Variants of Concern (VOC), unequal access to the vaccine supply and locally specific logistical and vaccine delivery parameters, add complexity to national CPH strategies and amplify the urgent need for effective CPH policies. Big Data and Artificial Intelligence Machine Learning techniques and collaborations can be instrumental in an accurate, timely, locally nuanced analysis of multiple data sources to inform CPH decision-making, vaccination strategies and their staged roll-out. The Africa-Canada Artificial Intelligence and Data Innovation Consortium (ACADIC) has been established to develop and employ machine learning techniques to design CPH strategies in Africa, which requires ongoing collaboration, testing and development to maximize the equity and effectiveness of COVID-19 related CPH interventions.

Evaluating temperature and humidity gradients of COVID-19 infection rates in light of Non-Pharmaceutical Interventions (preprint)

We evaluate potential temperature and humidity impact on the infection rate of COVID-19 with a data up to June 10th 2020, which comprises a large geographical footprint. It is critical to analyse data from different countries or regions at similar stages of the pandemic in order to avoid picking up false gradients. The degree of severity of NPIs is found to be a good gauge of the stage of the pandemic for individual countries. Data points are classified according to the stringency index of the NPIs in order to ensure that comparisons between countries are made on equal footing. We find that temperature and relative humidity gradients do not significantly deviate from the zero-gradient hypothesis. Upper limits on the absolute value of the gradients are set. The procedure chosen here yields 6{middle dot}10^-3{degrees}C^-1 and 3.3{middle dot}10^-3(%)^-1 upper limits on the absolute values of the temperature and relative humidity gradients, respectively, with a 95% Confidence Level. These findings do not preclude existence of seasonal effects and are indicative that these are likely to be nuanced.

Risk Adjusted Non-Pharmaceutical Interventions for the Management of COVID-19 in South Africa (preprint)

A global analysis of the impact of non-pharmaceutical interventions (NPIs) on the dynamics of the spread of the COVID-19 indicates that these can be classified using the stringency index proposed by the Oxford COVID-19 Government Response Tracker(OxCGRT) team. The world average for the coefficient that linearises the level of transmission with respect to the OxCGRT stringency index is s= 0.01{+/-}0.0017 (95%C.I.). The corresponding South African coefficient is s= 0.0078{+/-}0.00036 (95%C.I.), compatible with the world average. Here, we implement the stringency index for the recently announced 5-tier regulatory alert system. Predictions are made for the spread of the virus for each alert level. Assuming constant rates of recovery and mortality, it is essential to increase s. For the system to remain sub-critical, the rate with which s increases should outpace that of the decrease of the stringency index. Monitoring of s becomes essential to controlling the post-lockdown phase. Data from the Gauteng province obtained in May 2020 has been used to re-calibrate the model, where s was found increase by 20% with respect to the period before lockdown. Predictions for the province are made in this light.

Worldwide Effectiveness of Various Non-Pharmaceutical Intervention Control Strategies on the Global COVID-19 Pandemic: A Linearised Control Model (preprint)

Background COVID-19 is a virus which has lead to a global pandemic. Worldwide, more than 130 countries have imposed severe restrictions, which form part of a set of non-pharmaceutical interventions (NPI)s. We aimed to quantify the country-specific effects of these NPIs and compare them using the Oxford COVID-19 Government Response Tracker (OxCGRT) stringency index, p, as a measure of NPI stringency. Methods We developed a dual latent/observable Susceptible Infected Recovered Deaths (SIRD) model and applied it on each of 22 countries and 25 states in the US using publicly available data. The observable model parameters were extracted using kernel functions. The regression of the transmission rate, {beta}, as a function of p in each locale was modeled through the intervention leverage, s, an initial transmission rate, {beta}0 and a typical adjustment time, br-1. Results The world average for the intervention leverage, s=0.01 (95% CI 0.0102 - 0.0112) had an ensemble standard deviation of 0.0017 (95% C.I. 0.0014 - 0.0021), strongly indicating a universal behavior. Discussion Our study indicates that removing NPIs too swiftly will result in the resurgence of the spread within one to two months, in alignment with the current WHO recommendations. Moreover, we have quantified and are able to predict the effect of various combinations of NPIs. There is a minimum NPI level, below which leads to resurgence of the outbreak (in the absence of pharmaceutical and clinical advances). For the epidemic to remain sub-critical, the rate with which the intervention leverage s increases should outpace that of the relaxation of NPIs.