Diabetic retinopathy (DR) is an eye disease involving retina that may develop as a complication of diabetes and can lead to significant vision impairment in its early stages and may later cause even the blindness. Diagnosis of DR is typically accomplished by analysing non-invasive fundus images. This paper introduces a new method for detecting DR, which comprises the extraction of Local binary patterns (LBP) and statistical features from the fundus images, followed by the training of Machine learning (ML) models. The performance of the introduced features for detecting the DR is evaluated on a benchmark dataset. It is found that the combination of LBP and statistical features achieves significantly encouraging detection performance on SVM, k-NN, ANN, and Logistic regression (LR) classifiers, as compared to using the features individually. Among these classifiers, the detection performance of SVM is quite encouraging than others.