A neural network based time series forecasting

In a time series model of forecasting, given a set of past data values say, d(1), d(2), .., d(m) where d(i) represents the data at the i^th time instant, 1≤i≤m, the problem of forecasting is to estimate d(m+1) or more generally d(m+τ) for a small integer τ The least mean square (LMS) algorithm is well recognized as the linear adaptive filter which has diverse applications such as seismology, biomedical engineering, radar, control systems, communication systems etc., whereas the weighted moving average model is well known for a short term time series forecasting. In this paper, we show that the LMS algorithm on a single layer perceptron can also be used for short term time series forecasting. By simulating the algorithms, we also show that the LMS algorithm behaves very similarly as the weighted moving average model in producing the predicted result.