A Novel Approach to Farm Weather Prediction with Hybrid CNN, LSTM, and Attention Mechanisms

This paper presents a comprehensive approach for forecasting weather parameters such as temperature, humidity, and wind speed, using ML and DL techniques. The dataset with weather data covering several years, with features including date, temperature, wind speed, humidity, and precipitation is collected. The data pre-processing involved converting the 'Date' column to a datetime format and extracting key features such as 'Month', 'Season', and 'Day' to capture seasonal and temporal patterns. These features were used for further analysis and modeling. Exploratory data analysis was then performed to visualize seasonal patterns, followed by correlation and trend analyses. For predictive modeling, a range of ML models were applied, including Random Forest Regressor, SVM, XGB, and KNN. These models were evaluated based on several performance metrics. To further enhance the predictive performance, a DL model LSTM applied. Later, novel hybrid models combining CNN with LSTM and CNN+LSTM+Attention architecture demonstrated significant improvements in performance. The hybrid models showed superior generalization capabilities and accuracy in predicting weather parameters compared to traditional models.