CNN has emerged as the de-facto standard for several machine learning
(ML) and computer vision applications. It is known for its classification and feature
extraction capabilities. Many ML techniques require separate handcrafted feature
extraction steps before classification, which are “sub-optimal” in nature. Unlike these,
CNN extracts “optimal” features directly from raw data, enabling it to enhance
classification accuracy. Two-dimensional CNN (2D-CNN) is the most common one,
where inputs to the CNNs are 2D in nature, such as images. Here, we used 1D-CNN for
data classification as we used 1D inputs. 1D-CNN has lower computational complexity
than 2D-CNN. Mainly for this, we preferred 1D-CNN over 2D-CNN. To demonstrate
the superiority of the proposed generic classifier, we compared its classification
accuracies with several other generic classifiers. We used 21 benchmark data sets from
the UCI machine learning repository to achieve this. Tests prove the superiority of the
proposed 1D-CNN-based generic classifier. Many 1D-CNN-based application-specific
classifiers are proposed in the literature, but the proposed classifier is applicable for
many types of tabular data i.e., it is a generic classifier.
Keywords: 1D-convolutional neural network, Data classification, Machine learning.