The purpose of fundus imaging is to examine the anomalies related to diseases that affect the eye. A fundus image plays a crucial role in the observation and detection of various ophthalmological diseases. The majority of earlier researchers have concentrated their approaches on the identification of individual diseases from fundus image. But, simultaneous detection of multi-disease from fundus image is still facing a great challenge. To the patient, there is a chance of having more than one disease while diagnosing either or both the eyes. So, it is planned to address this challenge by designing a framework which can detect multi-disease from fundus image with improved accuracy. This paper proposes a novel Multi-Disease Classification Framework (MDCF) by incorporating ensemble neural architectures. In the proposed framework, the initial task is to perform preprocessing on the dataset with certain steps like: contrast enhancement, oversampling, resizing, and normalization. The MDCF will be conceded in two stages: the first stage is to detect whether the fundus image is at disease risk or not and the second stage is to classify multi-disease on fundus image. Two convolutional neural networks Densenet201 and EfficientNetB4 were used for disease risk detection and in addition to these two networks ResNet105 is added for multi-disease classification. Retinal Fundus Multi-disease Image Dataset (RFMiD) is used for training and validation of the proposed work. The MDCF is tested on Ocular Disease Intelligent Recognition (ODIR) 2019 dataset and the output demonstrated that the proposed work is performing well compared to the other stateof-the-art results.