Weather Forecasting Using Neural Networks with Backpropagation and ADAM Optimizer for city of Lhokseumawe

Abstract

Weather forecasting in Lhokseumawe is crucial due to its diverse climate and impact on community activities.
It serves as an operational responsibility of the Meteorology, Climatology and Geophysics Agency (BMKG) worldwide.
The method of forecasting currently employed by the BMKG involves meteorological teams observing and analyzing
statistics based on principles of mechanics and physics. Artificial Neural Networks (ANN) can be utilized to forecast
long-term weather conditions, with the backpropagation algorithm being an ANN algorithm employed for short-term
weather prediction. This involves training the backpropagation architecture data, which includes an input layer with a
size of 6 using Relu activation, one hidden layer with a size of 64 using Relu activation, and an output layer with a size of
3 using softmax activation. We also apply the ADAM optimizer, loss sparse categorical crossentropy, and accuracy
metrics. However, the backpropagation algorithm displays weaknesses, including slow convergence, overfitting, and
susceptibility to local minima, which can be addressed by utilizing the ADAM optimization algorithm. The research
utilizes Artificial Neural Network (ANN) with the backpropagation algorithm and ADAM optimization to predict
weather conditions in Lhokseumawe City with high accuracy. The research methods comprise of data collection,
preprocessing, division, model building, and evaluation. The study outcomes present the weather conditions as sunny,
cloudy, or rainy with an algorithm accuracy of 72%.