Quality measurement classification for water treatment using neural network with reinforcement programming for weighting optimization

Water Quality is a basic need for human. If the quality level of the water is not appropriate, it will give dangerous impacts to the human life. Therefore, the measurement of the water quality becomes important because it needs specific treatments to make more acceptable for specific uses of the water such as drinkable water, paddy fields, river flow maintenance, water recreation or other environmental preservation purposes. in this paper we present measurement classification of water quality for treatment with involving some parameters of the water such as Biological Oxygen Demand, Chemical Oxygen Demand, Ph, Suspended Solid, etc. We use Neural Network to deal with the classification problems and make classification into 13 types for the water treatment. In this paper we propose our Reinforcement Programming algorithm to optimize weighting mechanism for Neural Network. Reinforcement Programming is an optimization algorithm derived from Reinforcement Learning, a new learning paradigm in machine learning that learns from the interaction with external environments to achieve a goal. Reinforcement Programming improved the Reinforcement Learning by shifting goal-based to function-based approach in order to solve the optimization problems in weighting mechanism of Neural Network. It updated the weights of Neural Network by implementing the exploitation and exploration of Neural Network weights, and then measured the differences of state values from a given state-function to assign a reward or punishment of the state. We applied our proposed Reinforcement Programming to optimize Neural Network weighting mechanism for water quality measurement classification and made series of experimental study with water quality treatment dataset provided by UCI Machine Learning Repository. To scrutinize the applicability of our proposed approach, we made performance comparison with common existing Neural. Neural weight update using Backpropagation. In the experimental results, our proposed Reinforcement Programming outperformed Backpropagation for weighting mechanism in precision and time.