Medical Data Imputation by Hybrid Optimized Fuzzy C-Means and Ant Lion Optimization Algorithm

This paper uses a nature-inspired optimization algorithm called Ant Lion Optimization (ALO) to optimize the parameters of an imputation method, Fuzzy c-means (FCM). FCM is severely affected by the selection of parameters c and $\mathbf{m}$, which control the cluster number and fuzziness variable, respectively. Without the appropriate parameters selection, the prediction provided by FCM will result in low accuracy imputation. The traditional trial and error approach, which combines the parameters, is not reliable and prone to error as the value of c can range from 2 to the number of data while the value of m can range from 1 to infinity. Thus, a hybrid with a reliable optimization algorithm that can escape local optima and have fast computational time is appropriate. As such, a hybrid with an ALO optimization algorithm is proposed, inspired by the antlion hunting mechanism in nature. The proposed hybrid FCM-ALO was tested on three medical datasets: fertility, diabetes, and heart disease. Five missing rates are also used to test the proposed method’s reliability in increasing the rate of missing values further. The experimental results show that FCM-ALO produced a low error rate in Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) in all the datasets in all missing rates. Numerical and comparative results show that the proposed hybrid optimized imputation method has good convergence ability and better imputation quality in low computational time.