Brain tumour is the second most common leading disease in the world. It reduce the survival rate of a patient. Many automated systems and classification algorithms are available to detect brain tumours using MR images. World Health Organization initiates a next step to classify the brain tumours based on molecular features which helps to identify the histological subgroups for better prediction system. In this paper, instead of considering MR images, risk factors are used to identify histological type of tumours. To identify the best risk factors two phase feature selection algorithm is used, which composed by enhanced filtrate feature selection algorithm in phase I to identify the dependency and iterative feature displacement algorithm in phase II to achieve high quality and dimensional optimal dataset. The selected risk factors are classified in two ways such as by morphology and histology using new algorithm called Adaptive Recursive Partition Analysis (ARPART) based on target feature. The selected risk factors are also analysed using random forest, support vector machine and linear regression model. The proposed ARPART algorithm enables classification of tumour patient into more homogeneous and prognostic groups for better diagnosis process. The main aim of ARPART is to produce a homogeneous terminal node. ARPART algorithm shows 99.93% of accuracy in Histological based and shows 98.19 % of accuracy in morphological based. The proposed algorithm outperforms than other classifier models.