Evaluating the degree of cloudiness using machine learning techniques based on different atmospheric conditions

Cloud cover, the percentage of the sky obscured by clouds, has significant effects on agriculture, global warming, and solar energy. This research has investigated machine learning techniques to accurately predict cloud cover in Bangladesh, comparing the effectiveness of various algorithms such as k-nearest neighbors, decision trees, Naïve Bayes, support vector machine, logistic regression, gradient boosting machine, and random forest using data from the Bangladesh Meteorological Department from 1980 to 2022. Validation is conducted using confusion matrix analysis, which rigorously assesses validity and identifies errors. The study presents a methodology for using training and test sets to objectively estimate performance metrics, including precision, sensitivity, accuracy, positive and negative predictive values, \(\varvec{F_1}\) ${\mathit{F}}_{\mathbf{1}}$ score, and Cohen’s kappa coefficient. Results are based on the analysis of training and test data sets. The mentioned machine learning techniques utilize cross-validation. Random forest shows the highest performance. Cross-validation adjusts the results, confirming the random forest’s top accuracy, kappa value, and \(\varvec{F_1}\) ${\mathit{F}}_{\mathbf{1}}$ score. Next in order, the support vector machine performs well. Among decision tree algorithms, it was discovered that the CART model performed better than C5.0. The Naïve Bayes algorithm performs the least well.