Is Neural Network a Reliable Forecaster on Earth?

Title: Is Neural Network a Reliable Forecaster on Earth?

Abstract: This research study investigates the performance of Multivariate Adaptive Regression Splines (MARS) and artificial neural networks in predicting one-month ahead rainfall. The analysis is based on 87 years of rainfall data from the Kerala state in India. The study compares the predictions generated by both models and evaluates their efficiency using actual rainfall data. The results suggest that MARS is a better forecasting tool compared to the neural network.

Main Research Question: Can MARS and artificial neural networks accurately predict one-month ahead rainfall using historical data?

Methodology: The study uses 87 years of rainfall data from the Kerala state in India as the basis for its analysis. Both MARS and artificial neural networks are trained using scaled conjugate gradient algorithms. The trained models are then used to generate predictions for one-month ahead rainfall. The performance of both models is evaluated by comparing the predicted outputs with the actual rainfall data.

Results: The simulation results indicate that MARS performs better than the neural network in predicting one-month ahead rainfall.

Implications: The findings suggest that MARS is a more reliable forecasting tool for predicting one-month ahead rainfall compared to artificial neural networks. This study provides valuable insights into the effectiveness of different prediction models and could potentially influence future research in climate prediction and forecasting.

Link to Article: https://arxiv.org/abs/0405012v1 Authors: arXiv ID: 0405012v1