When talking about complicated machine learning systems, it is usual to assume that deep neural networks with plenty of parameters are the way to go. On the other hand, over fitting is a common problem with big network systems, which can cause poor performance. Consequently, effectively managing problems like over fitting becomes challenging. Dropout is the most effective strategy for addressing such difficulties. Over the course of the training process, the notion entails removing units and their connections from the neural network in a random fashion. This makes it difficult for units to co-adapt to an extreme degree. Throughout the training process, dropout will select samples at random from a collection of thinning networks. By utilising a single network with reduced weights. This technique mitigates over fitting and yields significant enhancements compared to alternative regularisation approaches. This study focuses on increasing the efficiency of systems in supervised learning applications such as vision, recognition of speech, text classification, and biological computation. This is to achieve state-of-the-art results on various benchmarked datasets.