Development of a Hybrid Adaptive Neuro Fuzzy Inference System - Harris Hawks Optimizer (ANFIS-HHO) Model to Predict Soil Cation Exchange Capacity

Abstract

Predicting the amount of soil exchange capacity is very valuable because it is a key indicator of soil quality for nutrient storage. In this research, using a neural-fuzzy network (ANFIS), the soil exchange capacity value was predicted with the input parameters of soil properties (such as clay, sludge, sand, gypsum and organic matter). The study area was considered in the northwest of Iran, where about 380 samples were taken from different areas. Of these samples, about 75% of the data were considered for training and 25% of the data were considered for testing. According to the number of different inputs, about 6 models of the combination of input parameters were developed. To improve the prediction results, the Harris hawk optimizer (HHO) algorithm was used for ANFIS training. The pattern results of each model were analyzed using evaluation indices such as root mean square error (RMSE), mean absolute percentage error (MAPE), correlation coefficient (CC), Williams index (WI), and Taylor diagram. The results showed that the input pattern P6, which includes all input parameters, had the highest accuracy in predicting CEC. In this pattern, the error values of CC, WI, MAPE and RMSE for the test data were 0.90, 0.75, 0.11 and 1.89 cmol /kg, respectively. The results of Taylor diagram also indicated the appropriate accuracy of the pattern so that the CEC can be predicted with appropriate accuracy. In general, the prediction error was reduced by about 1.3 to 2 cmol/kg using the HHO algorithm.

Keywords

• cation exchange capacity
• harris hawks algorithm
• neural-fuzzy network
• prediction