The prediction of early childhood numeracy skills development is often studied by determining the learner’s performance in a numeracy test. It is an important study area since numeracy impacts on the learner’s mathematical and statistical abilities later in life. Despite having pros and cons over each other, classification algorithms are often applied in the prediction of early childhood numeracy skills development without justifying the choice of a certain algorithm over others. In this paper, the bi-directional stepwise logistic regression model (SLRM), hierarchical logistic regression model (HLRM), classification and regression tree (CART) and Naïve Bayes (NB) algorithm were compared in terms of their ability to accurately classify some learners as having passed or having failed a numeracy test. The intention was to determine the most accurate algorithm from the four relative to predicting the learner’s performance in a numeracy test. The algorithms were compared using the true positive rate, true negative rate, specificity, sensitivity, classification error, classification accuracy and the area under the receiver operating characteristic (AUROC) curve. The results showed that the HLRM which has been applied by several previous studies on the prediction of numeracy test competence is the best classifier followed by SLRM, CART then NB. The study also confirmed some important predictors of the learner’s performance in a numeracy test some of which were also identified by some previous studies on early childhood numeracy development. Some gaps and recommendations for future research pertaining to the classification algorithms as well as implications for practice were also highlighted. We have made the HLRM scoring algorithm generated from SPSS available as a supplementary material and can be used to classify a set of new learners to either the pass or fail group.
Despite the growing criminal activities in South Africa, many victims still do not report the crimes, therefore there was a need to understand the determinants of the likelihood of reporting a crime in the country. Binary logistic regression is a supervised machine learning algorithm that can assist in predicting the likelihood of reporting a crime but the selection of relevant variables to add in the model varies from one author to the other. Selection of theoretically sound and statistically relevant independent variables is key to achieving parsimonious multivariate models. This study sought to test the efficiency of some commonly used variable selection methods for logistic regression models in order to identify the most relevant determinants of the likelihood of reporting a crime of housebreaking. The study used 17 candidate variables such as the victims’ demographic variables and their perceptions on the police. The multivariate model fitted using stepwise selection was found to be a best fit for the data based on the lowest AIC, the highest classification accuracy rate and the highest Area under the Receiver Operating Characteristic curve. The model fitted using the Hosmer-Lemeshow (H-L) algorithm was the worst fit for the data. The study revealed a limitation of the stepwise selection method which is that this method may select different independent variables for each unique set of randomly selected observations of the same dataset. The study established a multivariate logistic regression model to predict the likelihood of a victim reporting a crime of housebreaking and the determinants thereof.
20 July 2017 in Journal of Economics and Behavioral Studies
Sample size requirements are common in many multivariate analysis techniques as one of the measures taken to ensure the robustness of such techniques, such requirements have not been of interest in the area of count data models. As such, this study investigated the effect of sample size on the efficiency of six commonly used count data models namely: Poisson regression model (PRM), Negative binomial regression model (NBRM), Zero-inflated Poisson (ZIP), Zero-inflated negative binomial (ZINB), Poisson Hurdle model (PHM) and Negative binomial hurdle model (NBHM). The data used in this study were sourced from Data First and were collected by Statistics South Africa through the Marriage and Divorce database. PRM, NBRM, ZIP, ZINB, PHM and NBHM were applied to ten randomly selected samples ranging from 4392 to 43916 and differing by 10% in size. The six models were compared using the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Vuong’s test for over-dispersion, McFadden RSQ, Mean Square Error (MSE) and Mean Absolute Deviation (MAD).The results revealed that generally, the Negative Binomial-based models outperformed Poisson-based models. However, the results did not reveal the effect of sample size variations on the efficiency of the models since there was no consistency in the change in AIC, BIC, Vuong’s test for over-dispersion, McFadden RSQ, MSE and MAD as the sample size increased.
This study aimed to compare RNN algorithms and select the best-performing one between the GRU and LSTM for forecasting South African adjusted closing gold prices. The study used weekly secondary data sourced from Yahoo Finance and partitioned into three regimes, pre-COVID-19, COVID-19, and post-COVID-19, as well as the overall sample. The results indicated that the GRU algorithm consistently outperformed the LSTM algorithm across all evaluation periods based on the selected metrics, except during the COVID-19 period, where LSTM exhibited slightly better performance. Consequently, the GRU algorithm was identified as the best-performing algorithm for the South African adjusted closing gold price series. The relative effectiveness of GRU and LSTM algorithms in financial time series forecasting was clarified by the results. By integrating GRU-based forecasts into development finance frameworks, stakeholders can strengthen resilience against global shocks, improve financial planning, and foster more stable pathways for economic development. The authors recommended that future studies explore the performance of the GRU and LSTM with other advanced algorithms like Transformer architectures, hybrid algorithms, or traditional statistical methods to further enhance the forecasting robustness.
This paper compared count series, time series and machine learning models to determine the best data type (between count data and continuous data) and the best model for forecasting the labour market variables. Data from Statistics South Africa’s Quarterly Labour Force Survey (2008–2021) was used to compare ARIMA, Poisson autoregressive (PAR), negative binomial regression (NBR), generalised Poisson regression (GPR), support vector regression (SVR) and the multilayer perceptron (MLP) regressor based on the MAE, RMSE, MAPE, MSE and the plots of the actual values versus the forecasts from the models. The study showed that the most accurate approach to estimating the future values of the labour market variables is to implement PAR to forecast the counts of people who are unemployed, employed, and non-economically active, and then use these forecasts to compute the future values of unemployment rate, absorption rate and labour force participation rate.