Weather prediction on Data streams using EM-fuzzy clustering with Kohonen's self-learning

Abstract

Now a day, all the people are connected with internet and social media and becoming big data
creators and users. From the data streams finding a useful information is a challenging task.
FORBODE performed the prediction through the dynamic truncated back propagation and
enhanced L1 regularization method that eliminates the vanishing gradient issue by selecting the back
propagation time steps regarding the error function and deterministic weight. It employs Particle
Swarm Optimization (PSO) method to select the optimal features through the penalty term (λ) that
helps to avert the over fitting of data even with the advent of recent data. To improve the accuracy of
the prediction this work proposes Kohonen’s self-organizing map (SOM) based EM- fuzzy clustering
method to improve the prediction accuracy. Both the work was compared by the parameters such as
Precision, RMSE, Recall, F-Measure and MCC. The SOM approach surpasses the Forecasting
Based On Dynamic truncated back propagation and Enhanced L1 regularization (FOREBODE)
approach by establishing good performance in predicting the values.