| Abstract: | Recently, there has been upsurge of interest in the area of data mining in knowledge discovery using machine learning techniques. Traditionally, such attempts have been on classification and prediction; by establishing a transformation from input data to output data under a given performance criteria by creating a function approximation. When applying a classification or regression, there is a dichotomy in choosing whether to use local vs. global characteristics of the input data in machine learning algorithms which can be divided as global learning and local learning [7]. The Nile river flood forecasting problem has been traditionally tackled using statistical linear techniques, such as ARMA and hybridized with genetic algorithms [1, 13]. Moreover, the particle swarm optimization applied to train the neural networks to measure its predictability on flood forecasting [14]. In this paper, we demonstrate the results of utilizing the effect of gradually increasing the locally learning model for prediction the Nile flood forecasting using different machine learning algorithms. The motivation for focusing using local modeling is to avoid negative interference exhibited by global models. Moreover, the results of using locally learning model are compared with the results of using global learning models (i.e., back-propagation algorithm).
|