Modelling and forecasting cash outflow-inflow using ARIMA-Feedforward Neural Network

Abstract

Time series data modelling is often done in the world of the economy. In this study, modelling and forecasting of cash outflows and inflows were carried out, especially at the Bank Indonesia Representative Office in Malang City. The amount of money going out and going in needs to be modelled and forecasted to estimate people's money needs in the next period. Forecasting and modelling time-series data that is often used is the Autoregressive Integrated Moving Average. However, the modelling can be said to be a simple time series modelling and cannot model non-linear models. One of the developments of machine learning modelling to predict time series data is a Feedforward Neural Network model which is used for forecasting cash outflows. ARIMA (1,1,1) model is used to forecast cash inflow. The FFNN model cannot be used for these data because the cash inflow does not meet the non-linearity assumption. Meanwhile, cash outflow uses the FFNN model with ARIMA input (2,1,0) and eight hidden layers. The cash outflow modelling input used is based on the best ARIMA model and the determination of the number of hidden layers is done by selecting the smallest MAPE and RMSE values.