Page 50 - FULL REPORT 30012024

2.6.2 Performance of Machine Learning Prediction Algorithm

i. Support Vector Machine (SVM)

In a study conducted by H. Tan (2021), that focused on text
classification tasks SVM was shown to have the lowest accuracy

among the classifiers employed. This implies that SVM might not be a

good fit for handling text data in that specific study. SVM finds it
difficult to identify the best decision boundary when dealing with text

data because of its frequent high dimensionality and sparsity. Other
classifiers, such as Naive Bayes may have fared better in this example

due to their capacity to handle the special properties of text input.

SVM also did not perform as well as the random forest classifier that

sought to identify paediatric asthma patients at risk of hospital
readmissions (Shin et al., 2018). This suggests that random forest was

more accurate in determining which asthma patients were most likely
to visit the hospital again. The precise causes of this performance

disparity may vary, but they could be traced to random forest's capacity

to capture complex connections and interactions between different
elements in the dataset. Random forest's ensemble nature allows it to

make predictions by combining the outputs of multiple decision trees,
which can be advantageous in cases where the underlying patterns are

not easily captured by a single SVM model.

ii. Naïve Bayes (NV)

Naive Bayes has demonstrated strong performance in several

circumstances, such as credit rating (Pal, 2020). NV was used in a
study conducted by Ginting et al. (2018) to analyse bank customer

data, and the accuracy was 94%. This high level of precision indicates
that NV was successful in correctly identifying bank clients based on

the supplied information.

33