<p>Diabetic retinopathy (DR) is a serious complication associated with diabetes, potentially causing substantial vision impairment if not identified and addressed promptly. The increasing prevalence of diabetes worldwide necessitates efficient and accurate methods for DR detection, especially in regions with a shortage of ophthalmologists. In this study, we propose a lightweight hybrid model that uses Deep Convolutional Neural Networks (DCNNs) as feature extractor. The proposed work is divided into two phases: Feature extraction and classification. The feature extraction process involves four different DCNN models: EfficientNetB6, VGG19, ResNet50, and InceptionV3. Subsequently, we have employed six different machine learning models, Random Forest, Support Vector Machine (SVM), eXtreme Gradient Boosting (XGBoost), K-Nearest Neighbour (KNN), Logistic Regression (LR), and Decision Tree (DT) as classifiers to categorize the images based on the extracted features. This work has been evaluated on the Asia Pacific Tele-Ophthalmology Society (APTOS) dataset. This dataset consists of three parts: training, testing and validation set. The training set is used for training the models, validation set images used for validating the results and final testing is done on the test set of the dataset. In contrast to the current practice, which mostly addresses later stages of DR, we will have an early detection strategy to use Class 0, 1, and 2 images to help stop vision loss in its early stages. All the models that were tested resulted in 96.55 percent as the highest accuracy score, owing to their use of InceptionV3 as a feature extraction model and the SVM to classify the results as the best model to detect early-stage DR.</p>