Kishana R. Kashwan and Velu C. Muni
Soft computing, information technologies, data mining, neural networks
The neural networks have been used and tried for many intelligent processes to mimic human decision making power. Classification technique is a vital step in data mining for intelligent applications. Neural networks are quite often trained much in similar way of human learning process. The efficacy of neural networks depends upon how precisely these are trained. Training can be of many types. We have chosen harmony search algorithm to train the neural networks for its known advantages. The scope of this paper is focused on to examine the supervised learning of neural networks by employing harmony search algorithm to predict market selling trends for different items in order to support decision making and planning. We gathered data sets from market transactions to train and test real time and online neural network model. The model predicts sales automatically and updates the records after every transaction. We have implemented and tested the model over a long period by comparing actual sales with predicted ones and taking cyclical and seasonal effects into account. Results are quite encouraging and have shown good accuracy. We have analysed model by comparing with existing alternative techniques. It shows very competitive and high classification accuracy.
Important Links:
Go Back
