Abstract:
This paper focused on the development of a predictive model
for the classification of the risk of kidney stones in Nigerian
using data mining techniques based on historical information
elicited about the risk of kidney stones among Nigerians.
Following the identification of the risk factors of kidney stone
from experienced endocrinologists, structured questionnaires
were used to collect information about the risk factors and the
associated risk of kidney stones from selected respondents.
The predictive model for the risk of kidney diseases was
formulated using three (3) supervised machine learning
algorithms (Decision Tree, Multi-layer perception and
Genetic Algorithm) following the identification of relevant
features. The predictive model was simulated using the
Waikato Environment for Knowledge Analysis (WEKA)
environment; and the model was validated using historical
dataset of kidney stone risk via performance metrics:
accuracy, true positive rate, precision and false positive rate.
The paper concluded that the multi-layer perceptron had the
best performance overall using the 33 initially identified
variables by the endocrinologists with an accuracy of 100%.
The performance of the genetic programming and multi-layer
perceptron algorithms used to formulate the predictive model
for the risk of kidney stones using the 6 variables
outperformed the model formulated using the 6 variables
identified by the C4.5 decision trees. The variables identified
by the C4.5 decision trees algorithm were: obese from
childhood, eating late at night, BMI class, family history of
hypertension, taking coffee and sweating daily. In
conclusion, the multi-layer perceptron algorithm is best
suitable for the development of a predictive model for the risk
of kidney stones.