Comparison of Neural and Bayesian Networks for Real-Time Data Classification
Neural and Bayesian networks have been successfully used in different classification tasks during the last several decades. Then during the last several years, the interest towards deep neural networks have been hugely increased and they have started to be used in vast majority of fields including image, speech, signal processing. Currently field researchers and specialists try to apply neural networks in almost every sphere and system, including systems that deal with real-time data. Eventually neural networks became more popular in industry than Bayesian networks. However, there are some concerns and unanswered questions about this type of usage of neural networks. Especially neural networks are being misused very often in classification tasks, and field specialists do not consider the fact that Bayesian networks could be better solution with better performance and accuracy for a specific problem. In addition, there is a need to consider some factors before choosing the network type, such as transparency of the algorithm, theoretical justification, missing values in data, restriction of being only supervised approach, network building and training time, adaptiveness in case of real-time data. In this work, we present differences of neural networks and Bayesian networks, more specifically for classification tasks for real-time data and carry out theoretical and practical comparison between them. Afterwards, we provide some ideas on which approach is preferable in case of real-time data classification.