The Fuzzy Min max Neural Network (FMNN) developed by Simpson is defined as a neural network that forms hyperboxes for classification and prediction. This paper proposes an improvement in learning algorithm in FMNN using semi-supervised clustering method, called SS-FMM. The proposed model combines the advantages of supervised learning and those of unsupervised learning. Labeled a part of data is the additional information that is used in this semi-supervised clustering method. For evaluation purpose, this algorithm is implemented on two datasets including Shape sets from CS and Thyorid disease from UCI. A part from that, in this paper, some related algorithms in FMNN are also setup on these datasets in order to compare the accuracy with proposed algorithm. The test results show that the novel algorithm has the better performance.