EVALUATING THE PERFORMANCE OF RANDOM FOREST AND MONTE CARLO SIMULATION FOR FORECASTING MONTHLY TEMPERATURE VARIATION IN NORTHERN MINDANAO PHILIPPINES

In this paper, we analyzed the historical temperature data spanning from 2010 to 2020 of Malaybalay City, Bukidnon using the random forest regression and Monte Carlo simulation. Such advanced methods were duly presented to enhance the accuracy of the monthly forecasting of temperatures. The results show that the random forest regression model has performed better, with an MSE of 0.67 and an RMSE of 0.82 on the test dataset, thus indicating that its temperature value predictions are closer to the real ones than ARIMA. However, ARIMA performed well on training data but had a much higher MSE of 3.03 and an RMSE of 1.74 on the test dataset, showing overfitting and less accuracy on the unseen data. The combined approach with random forest regression and Monte Carlo simulation returned an MSE of 0.68 and an RMSE of 0.83 against the test dataset, performing slightly better than ARIMA and behind the random forest model. These results further support how well random forest regression can model complex, nonlinear interactions, handle anomalies in various temperature data, and were further proved to be very effective in predicting temperatures.