Predictive Maintenance for IoT-Enabled Wireless Devices Using AdaBoost with Synthetic Minority Oversampling Technique

Nowadays, the rapid growth of predictive maintenance methods has become crucial for Internet of Things (IoT) enabled wireless devices. The existing Random Forest (RF) utilized to predict device failures although, the complexity of IoT device data led to scalability issues. Hence, this research proposes an Adaptive Boosting with Synthetic Minority Oversampling Technique (AdaBoost-SMOTE) for predicting device failures. The proposed AdaBoost-SMOTE overcome the scalability issues by handling imbalanced data and reducing dimensionality for accurate prediction. Initially, the input data is collected from IoT-enabled sensors and preprocessed with z-score normalization to scale the features. Then, Permutation Importance (PI) is employed to select the optimal features by ranking the feature score based on its importance. After that, SMOTE is used to handle the imbalance data and then, AdaBoost is employed to predict the machine failures by assigning the higher weights to misclassified samples in an iterative manner. Finally, hyperparameter tuning is done by using Grid Search Cross-Validation (GSCV) to optimize the parameters by selecting best CV score as a best set of hyperparameters. From the results, the proposed AdaBoost-SMOTE achieved best results in terms of accuracy (97%), precision (95%), recall (93%), and F1-score (94%) when compared to existing AdaBoost.