Abstract:
Acute Flaccid Paralysis (AFP) is caused due to diseases like Poliomyelitis, Acute
Transverse Myelitis (ATM), Guillain-Barre-Syndrome (GBS), Traumatic Neuritis (TrN), and
others of similar nature. These diseases can damage various muscles or spinal cord, or
peripheral nerves. Pakistan is among the few countries which are struggling to eradicate
Poliomyelitis. The conventional diagnosis takes a lot of effort and time. A solution to this
problem is provided in this research using the methods from Data Science. Objective:
Diagnosis of AFP is done via two approaches, used by the framework proposed in this
research: 1) Prediction of diagnosis for non-polio AFP cases, using patient’s dataset; 2)
Prediction of clinical diagnosis for AFP, using the developed rule base data based on a
standard chart, provided by the World Health Organization (WHO) for differential
diagnosis of AFP. Method: 1) The laboratory diagnosis method predicts the diagnosis of
non-polio AFP cases using patient data. The data was preprocessed and converted into a
dataset; which was used for model training and prediction of diagnosis for non-polio AFP.
2) The clinical diagnosis method is based on the symptoms given in the standard table of
WHO. This table differentiates between diseases like Poliomyelitis, GBS, TrN, TM, and
Acute Childhood Hemiplegia. A rule base was devised according to this standard diagnosis
table, with the five diseases acting as labels of the dataset. The rules helped in diagnosis for
a particular AFP case. Decision Tree and Random Forest are the two algorithms used for
model training and prediction in this research. The trained model predicts the disease for a
particular AFP case. Results: For the first method, the prediction accuracy was 87.76%
with the decision tree algorithm and 92.46% with random forest. For the second dataset,
using the rule base dataset, the accuracy with the decision tree algorithm was 98.54% and
100% with random forest.
