Search

User churn is an important issue in online services that threatens the health and profitability of services. Most of the previous works on churn prediction convert the problem into a binary classification task where the users are labeled as churned and non-churned. More recently, some works have tried to convert the user churn prediction problem into the prediction of user return time. In this approach which is more realistic in real world online services, at each time-step the model predicts the user return time instead of predicting a churn label. However, the previous works in this category suffer from lack of generality and require high computational complexity. In this paper, we introduce ChOracle, an oracle that predicts the user churn by modeling the user return times to service by utilizing a combination of Temporal Point Processes and Recurrent Neural Networks. Moreover, we incorporate latent variables into the proposed recurrent neural network to model the latent user loyalty to the system. We also develop an efficient approximate variational inference algorithm for learning parameters of the proposed RNN by using back propagation through time. Finally, we demonstrate the superior performance of ChOracle on a wide variety of real world datasets. [ABSTRACT FROM AUTHOR]

Modularity as a system of separate and independent sub-tasks is the appropriate way to improve the performance of artificial neural network (ANN) models in hydrological processes. Using this approach, a block combined neural network (BCNN) structure incorporated with genetic algorithm (GA) and an additional decision block is suggested in this study. The optimum topology of embedded networks in each block was detected using a vector-based method subjected to different internal characteristics. This model was then applied on 879 bedload datasets, considering velocity, discharge, mean grain size, slope, and depth as model inputs over streams in Idaho, USA. The correct classification rate of predicted bedload using BCNN (89.77%) showed superior performance accuracy compared to other ANNs, and to empirical models. Results of computed error metrics and confusion matrixes also demonstrated outstanding progress in BCNN relative to other models. We show that BCNN as a new method with an appropriate accuracy level could effectively be adopted for bedload prediction purposes. [ABSTRACT FROM AUTHOR]

Modeling and prediction of bed loads is an important but difficult issue in river engineering. The introduced empirical equations due to restricted applicability even in similar conditions provide different accuracies with each other and measured data. In this paper, three different artificial neural networks (ANNs) including multilayer percepterons, radial based function (RBF), and generalized feed forward neural network using five dominant parameters of bed load transport formulas for the Main Fork Red River in Idaho-USA were developed. The optimum models were found through 102 data sets of flow discharge, flow velocity, water surface slopes, flow depth, and mean grain size. The deficiency of empirical equations for this river by conducted comparison between measured and predicted values was approved where the ANN models presented more consistence and closer estimation to observed data. The coefficient of determination between measured and predicted values for empirical equations varied from 0.10 to 0.21 against the 0.93 to 0.98 in ANN models. The accuracy performance of all models was evaluated and interpreted using different statistical error criteria, analytical graphs and confusion matrixes. Although the ANN models predicted compatible outputs but the RBF with 79% correct classification rate corresponding to 0.191 network error was outperform than others. [ABSTRACT FROM AUTHOR]