Zeitschrift für Infektionskrankheiten und Präventivmedizin
Offener Zugang
ISSN: 2329-8731
ISSN: 2329-8731
Azamossadat Hosseini, Hamid Moghaddasi, Reza Rabiei, Sara Mohebi Mushaei
Background: Data mining techniques for disease diagnosis help the prediction and control of various diseases, including Tuberculosis (TB). This study aimed to compare the efficiency of two main models of TB diagnosis: MLP (Multilayer Perceptron Neural Network) and ANFIS (Adaptive Neuro-Fuzzy Inference System) to find out which data- mining-based model is more efficient in detecting tuberculosis.
Materials and methods: In this analytical study, database used was for inpatients in a specialized hospital for lung and respiratory diseases. The database included 1159 records, of which 599 records belonged to TB infected patients and 560 records to non-infected patients. With help of 13 factors effective on diagnosis of the disease and using the set of TB records, the two models of MLP and ANFIS were tested and evaluated. Finally, using the ratio test, two models were compared based on their AUC values to see which one is more efficient. The sensitivity, specificity, accuracy, and RMSE of the two models were also compared.
Results: The efficiency of MLP was 0.9921 and the efficiency of ANFIS was 0.8572. MLP’s sensitivity, specificity, accuracy, and RMSE were recorded as 93.50%, 94.80%, 94.30%, and 0.1788, respectively. These values for ANFIS equaled 79.60%, 92.60%, 85.63%, and 0.3345, respectively. According to these results, there was a significant difference between efficiency levels of MLP and ANFIS models (p-value˂0.0001).
Conclusion: The MLP indicated a higher AUC value compared with ANFIS. The results also showed higher sensitivity, specificity, and accuracy but lower RMSE for MLP. Overall, MLP proved superior to ANFIS for TB diagnosis.