Combination of Adaptive Fuzzy Inference System and Simulated Annealing Algorithm-based for Malaria Susceptibility Mapping in Daknong Province

Abstract

Adaptive Neuro-Inference system (Anfis) has been widely used in recent studies aiming at generating probabilities of unseen data in binary classification application. It is normally used in combination with optimization algorithms for tuning its parameters to generate optimal objective values. This study proposed a state-of-the-art method using Simulated Annealing to improve Anfis performance. Malaria occurrences and spatial variation of environmental, socio-economic factors in Daknong province, Vietnam were selected for case study. For accuracy assessment, Receiver Operating Characteristic curve, Cost curve were used and the predicted map was compared to several benchmark classifiers. The results showed that the S-Anfis (AUC = 0.912, RMSE =0.335) outperformed Support Vector Machine (AUC = 0.902, RMSE =0.364), Multiple Layer Perceptron (AUC = 0.868, RMSE =0.430). Although, the performance of S-Anfis depended on proper selection of input factors and geographic variations of those, we concluded that this method could be an alternative in mapping susceptibility of malaria