A hybrid approach for data classification based on mathematical modelling and improved online learning algorithm for general fuzzy min-max neural network

The Fuzzy Min-Max Neural Network (FMNN) is a machine learning algorithm that utilizes hyper-boxes for classification and clustering tasks. It is built using hyper-box fuzzy sets, which allow for the representation of fuzzy data. This makes FMNN particularly effective for handling uncertain and imprecise data. Both the FMNN classification and clustering are based on this concept, making it a powerful tool for a variety of applications. Mixed integer programming models can offer improved performance and global optimality compared to other methods, but they are limited in their ability to handle large datasets because of the presence of binary variables and can be difficult to solve optimally. In this thesis, we aim to enhance the MILP model proposed by a well-known publication by providing an initial solution. This thesis presents a new hybrid approach that integrates mixed integer linear programming (MILP) and an improved online learning algorithm for general FMNN (IOL_GFMM). The IOL_GFMM method is employed to generate initial hyper-boxes to enhance the efficiency of the MILP model. The new hybrid approach is tested on both real and artificial datasets and has been shown to be efficient. The new hybrid approach significantly speeds up training time by up to 76 times and reduces the number of misclassified data by up to 17 times.