Rainfall time-series forecasting is an important research area which has applications in several fields like flood forecasting, drought prediction, water resource planning and management, precision agriculture and disaster management, to name a few. This paper discusses about a machine learning method called the Extreme Learning Machine (ELM) for predicting rainfall. The study area is Coonoor region, Tamil Nadu, India, which is prone to rainfall induced landslides. Two data sets have been used in this study. Data set 1 comprises of daily rainfall data of Coonoor, meteorological parameters like temperature, wind speed, relative humidity cloud cover and month, for the period 2004-2013. Data set 2 consists rainfall data of 14 rain gauge stations and month. A comparative study between the data sets is performed to show that only rainfall data is sufficient to accurately predict the rainfall in the given region. This is substantiated by performing sensitivity analysis on both the data sets. Sensitivity analysis also provides the most important predictor that contributes to accurate prediction of rainfall.
Improved Extreme Learning Machine based on the Sensitivity Analysis
