Analysis of Climatic Factors and Utilization of Machine Learning Techniques to Anticipate Humidity Levels in Northern Bangladesh

Analyzing meteorological data in the northern region of Bangladesh is crucial for understanding various aspects influenced by humidity. This study employs machine learning algorithms, including k-nearest neighbor, Classification and Regression Trees, C5.0, Naive Bayes, Random Forest, and Support Vector Machine, to forecast the humidity of northern Bangladesh. Data from 1981 to 2020 from two meteorological stations, Rangpur and Dinajpur, were utilized. Results indicate that Rangpur had the highest average daily humidity (80.34%), while Dinajpur had the lowest (77.26%). Cloud amount correlates positively with humidity and inversely with temperature. The k-nearest neighbor, random forest, and support vector machine algorithms generally revealed better prediction performance than other algorithms. All things considered, the Random Forest model demonstrates superior performance on the testing dataset at both stations, achieving 70% accuracy, <I>F<sub>1</sub></I>-score (75.85%), and a kappa value of approximately 53.3% at Rangpur Station, and 74% accuracy, <I>F<sub>1</sub></I>-score (78.4%), and a kappa value of approximately 60% at Dinajpur Station. Subsequently, this study analyzes the best performance and accuracy of the random forest classification algorithms through k-fold cross-validation for predicting humidity. With this piece of information, it is anticipated that the study underscores the importance of random forest in predicting humidity and aiding decision-makers in water demand management, ecological balance, and health quality in the northern region of Bangladesh.