Document classification is the task to split the document set into distinct highly relative classes or groups based on nature of the document contents.Here, an improved approach of document classification called keywordbased document classification (KBDC) is introduced. It focuses on splitting the unstructured text document set into K number of dissimilar classes based on K predetermined keywords text models by improved probability technique. This new system comprises of the following stages. Namely, pre-processing, classification and classifier stage respectively. Initial, the proposed system (KBDC) recognizes all the immaterial existing contents in the input text document through constructed Predetermined Irrelevant Text Pattern Model (PITPM). Next, it divides the pre-processed document set into 'K' different groups or classes by K number of Predetermined Keyword Text Pattern Models (PKTPM) through probability technique, where K denotes the number of groups or classes or models. Finally, the KBDC system classifies the trial test text document without any class label that belongs to either of the existing group based on the K different class models (PKTPs). Experimentation results show that the KBDC is appropriate to split and identifies the unstructured text document set into K distinct extremely comparative classes.