Enhanced Weather Forecasting Using Random Forest, LSTM, Gradient Boosting, and Ensemble Learning (RF + GB)

In the weather forecasting system, prediction is the most challenging process that involves the sequential data analysis to identify the most expressive pattern and to create a higher number of accuracy levels. This work puts forward a new approach for designing a weather prediction system utilizing LSTM networks that is a type of deep learning networks able to model temporal dependencies in the time-series data. This research involves data preparation for analysis and response formulation, regarding the given multi-attribute weather dataset. Such transformations include: how to deal with missing values, how to create new features and normalize the inputs for enhanced predictive capability. In this paper, the relevant parameters of severe weather forecasts are temperature, humidity, and precipitation, which are set as the targets for LSTM model. The outcomes analyzed in the experiment indicate that LSTM model is better in prediction, and the values of MSE and MAE are lower than the values obtained in other similar researches. Such results may demonstrate that LSTM networks have the potential that enhance the current weather forecast, so far, will be of great help in agriculture, transportation logistics, and disaster relief plans field.