A Triangular based determination of temperature using artificial intelligence

The forecasting method emerged in the middle of the twentieth century; its usage has grown exponentially in all aspects of life. More importantly, estimating modern meteorological parameters helps make good decisions regarding weather, health, and agricultural safety measures. Similarly, this study aims to find a better-fitting technique to translate Quetta's (Pakistan) temperature distribution using its three neighboring stations, Chaman, Kalat, and Sibi. In this regard, a well-known machine learning technique named Artificial Neural Network was utilized. Additionally, four training algorithms are also considered to optimize the model performance. Apart from that, another traditional statistical model is incorporated, which is a Multiple Linear Regression (MLR). Since the temperature distribution has a nonlinear trend, MLR techniques are also useful for making predictions. Machine learning and linear statistical models are provided with seven years of data from 2011 to 2017 for training purposes. Three sets of data for 2018, 2019, and 2020 are fed to determine how these trained models show close agreements with the actual temperature distribution. Different errors are evaluated to assess model performance, such as mean squared error (MSE), mean absolute percentage error (MAPE), mean absolute bias error (MABE), and chi-squared error. chi 2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\chi }<^>{2}$$\end{document}, and coefficient of determination (R2). For ANN, the models with the lowest MABE and MAPE values are ANN-RB and ANN-BR, whereas the model with the lowest MSE value, 1.3604, is the ANN-BFG model. The model with the highest correlation is the ANN-BFG model. On the other hand, MLR has an MSE of 1.4253 and a coefficient of determination of 0.9860.