Diabetes prevalence in recent years has increased diabetes complications like vision loss, skin diseases, heart, and kidney issues. The alarming complication is diabetic retinopathy (DR) if not detected at an early stage can result in permanent vision loss. Traditional techniques for DR detection rely on comprehensive image analysis, which results in drawn-out diagnostic procedures. DR detection needs to represent images in such a way that the processing along with computational cost can be reduced. The proposed work aims to perform a detailed survey on identifying various methods to reduce the computational cost. The outcome of the proposed work is that by identifying the most relevant and independent features in diabetic retinopathy so that the accuracy can be improved. The computational cost can be reduced by having optimal feature selection and extraction techniques as part of data preprocessing in Machine Learning (ML) and Deep Learning (DL) approaches.