Predictive Maintenance
1 oz. Prevention = 1 Lb. Cure…
Renewable energy is the way of the future and like any new technology, there will be growing pains… This supervised learning model uses data from a variety of sensors located on the turbine to monitor individual part performance and predict failure. Maintenance is a lot cheaper than replacement. By replacing parts before catastrophic failure occurs, we can improve the viability of the industry as a whole by reducing operating cost.
EDA
Data was scaled and missing values imputed with KNN and MICE
Bagging, Random Forest, Gradient Boost, Adaboost, XGBoost and Decision Tree models were evaluated on original data with Cross Validation
Data was treated for bias with Oversampling and Undersampling
Hyperparameter Tuning with Gridsearch and Randomsearch was applied to optimize model performance
Final Pipeline for deployment
