This research focuses on the development of predictive maintenance (PdM) systems using the Internet of things (IoT) for data extraction. Unsupervised algorithm is used to recognize and remove anomalies to ensure accuracy and reliability. Principal component analysis (PCA) is used as a preprocessing approach to efficiently reduce spatiality of dataset while preserving the most important information. Symmetric minority over-sampling technique is used as an oversampling technique, and MinMax scaler is used for normalization. Next, the study looks at models consisting of supervised algorithms which are trained and evaluated on the preprocessed dataset using binary classifiers and are evaluated for their ability to predict maintenance needs. The model’s accuracy is used as a key metric to measure model reliability in real-world scenarios. The dense neural network (DNN) model is used to target anomalies to evaluate model accuracy and loss. We have evaluated our machine learning techniques using precision and recall scores, and the average of the results for accuracy in binary classification test scores and validation scores are 92.05% and 91.20%, respectively. For multiclass classification, test scores and validation scores are 94.30% and 92.56%, respectively. Moreover, for deep learning techniques, accuracy score stands at 93.20%, and loss stands at 16.75%.