查看文章

A dataset may consist of hundreds, thousands, or millions of features which represent the whole data. Greater the number of features in the dataset, higher will be the complexity of data analysis process, thereby increasing the time and space complexity of the algorithm. One of the possible solutions to reduce the complexity of analysis process is to use dimensionality reduction technique which helps in minimizing the complexity of an algorithm. Dimensionality reduction is an essential activity performed prior to any data analysis process to reduce number of features from the dataset. In this experimental study, convolutional autoencoder has been implemented to study the impact of kernel size and activation function on the accuracy of algorithm. From the experimental results, it can be concluded that (3 * 3) is the best choice for kernel size and PReLU is best suitable for activation function used in the convolutional layers.