The implementation utilized a comprehensive weather forecasting dataset from Kaggle containing daily meteorological observations with multiple features including temperature, humidity, atmospheric pressure. The target variable represents binary weather conditions, distinguishing between rainy and non-rainy days. The dataset exhibits the typical characteristics of real-world weather data, including natural class imbalance where rainy days constitute a minority of the total observations. The data was subjected to preprocessing and other feature engineering techniques to improve prediction accuracy. The preprocessing pipeline began with comprehensive data quality assessment, including identification and handling of missing values, outliers, and inconsistencies in the meteorological measurements. Categorical variables were properly encoded, with the primary target variable (rain/no rain) converted to binary format (1 for rain, 0 for no rain) to facilitate the subsequent modeling process. Feature engineering was considered but kept minimal to maintain interpretability and focus on the core methodological contributions of the study. \subsection{Train-Test Split Strategy} A critical aspect of the methodology involved a sequential train-test split to preserve the temporal nature of the data. The dataset was chronologically ordered, and the first 80\% of the data was allocated for training, with the subsequent 20\% reserved for testing. This approach ensures that the model is trained on past data and evaluated on future, unseen data, simulating a real-world forecasting scenario. To address the class imbalance problem, a strategic oversampling approach was implemented exclusively on the training data after the train-test split. This approach involved duplicating instances of the minority class (rainy days) to create a more balanced distribution for model training. The duplicated values were concatenated with the training data and shuffled thoroughly. The oversampling ratio was determined empirically to achieve reasonable balance while avoiding excessive duplication that might lead to overfitting. \subsection{Feature Scaling} Given the continuous nature of meteorological variables and their different scales and units, feature scaling was implemented using StandardScaler from scikit-learn. The same scaling parameters derived from the training data were then applied to both training and test sets to ensure consistent preprocessing. The core of the methodology involved implementing a Gaussian Hidden Markov Model using the hmmlearn library. The GaussianHMM was selected because the input features consist of continuous meteorological variables that can be reasonably modeled using Gaussian emission distributions.