Introduction

This project aims to predict equipment failures using machine learning techniques. The goal is to help in preventive maintenance by identifying potential failures before they occur, thereby reducing downtime and maintenance costs.

Dataset

The dataset used in this project contains various metrics collected from different devices. Each entry in the dataset represents a snapshot of the device's state at a given time, including whether a failure occurred.

Data Preprocessing

Steps Involved:

• Data Cleaning: Remove redundant and irrelevant columns.
• Data Transformation: Convert categorical variables into dummy variables and handle skewness in numerical variables using log transformation.
• Feature Scaling: Apply standardization to numerical features.
• Feature Selection: Select important features using feature importance scores from a tree-based model.

Model Training

The following models were trained on the preprocessed data:

• Logistic Regression
• Random Forest Classifier
• XGBoost Classifier

Train-Test Split

The data was split into training and testing sets using an 75-25 split.

Hyperparameter Tuning

GridSearchCV was used to perform hyperparameter tuning for the Random Forest and XGBoost models to find the best set of parameters.

Evaluation

The performance of the models was evaluated using the following metrics:

• Precision
• Recall
• F1-Score
• ROC-AUC ROC curves were plotted to visualize the performance of each model.

Conclusion

The tuned models were able to effectively predict equipment failures, with the XGBoost model showing the best performance. By identifying potential failures early, this project can help in scheduling timely maintenance and reducing downtime.