Ensemble Methods based Machine Learning Approach for Weather Prediction for Precision Agriculture

Agriculture sector in developing countries including Sri Lanka has been highly affected by recent climatic variabilities. Expected raining might not occur at the expected time, due to recent changes happened in the rainfall patterns. This research mainly addresses the problems occurred due to the difficulty in identifying the mismatch between current rainfall and traditional seasonal cultivation schedules. Even with the advance technologies, weather prediction has become challenging task. This paper presents an Ensemble method based machine learning approach for predicting rainfall for precision agriculture. For predicting daily rainfall occurrence, six correlated weather features including wind speed, wind direction, humidity, temperature, O3 concentration and solar radiation were selected and improved with proper data preprocessing. 80-20 cross validated data were initially trained and tested with four ML algorithms where SVM, and Random Forest, KNN and Decision Tree achieved 88%, 89%, 89% and 90% accuracy levels respectively. Then several ensemble methods including Averaging, Voting and Bagging were used, where accuracy levels were 92%, 89% and 89%. Boosting method which combined AdaBoost with Decision Tree recognized as the best fitted model with 94.12% accuracy. The results depict a significant accuracy in this Ensemble methods based approach for Sri Lankan weather prediction