Accurate diagnosis of tumour type is a paramount issue in the appropriate treatment of cancer. In this study, a novel multi-step method, recognising benign and malignant tumour slices in real brain computed tomography (CT) images is proposed. First, image reconstruction is conducted for mixed noise removal using ‘weighted encoding with sparse non-local regularisation’ technique. For the segmentation purpose, support vector machine (SVM) is performed on CT images. Afterwards, 17 features are extracted, among which multiple important features are selected by the genetic algorithm. These selected features are used as the input to the multilayer perceptron neural network, the weighted kernel width SVM and the k-nearest neighbours models in tumour classification phase. Following this, the outcomes of mentioned classifiers are combined by means of multi-objective differential evolution-based ensemble technique, in order to enhance the classification performance indices. The parameters of this weighted linear combination are found by solving an optimisation problem containing precision and recall as the objective functions. The performance of the implemented approach is compared with the experienced radiologist ground truth and some state-of-the-art methods. The results demonstrate that the ensemble technique achieves high classification rates including the accuracy of 98.65%.