Predicting Machine Failure using Machine Learning Classification Models

The goal of this project is to use historical data of machines and their failures to predict whether a machine is likely to fail or not.

Motivation

Machine failure can be costly for companies in terms of downtime, maintenance, and repair costs. Predictive maintenance has become an important tool for companies to reduce these costs by identifying potential failures before they occur. Machine learning classification models are well-suited for predicting machine failure, and can help companies implement proactive maintenance strategies.

Data Source

The data used for this project is from Kaggle. The dataset contains information on various machines and their failures, including attributes such as Temperature, Type, Rotational Speed, Torque and Tool wear in minutes.

EDA (Exploratory Data Analysis)

Checking distribution and correlations.

Data Preprocessing

The data is preprocessed to handle outliers values and remove irrelevant features.

Data Pipeline

Pipeline to convert categorical features into numbers and testing MinMaxScaler and Standard Scaler.

Cross-Validation of models

We compared several classification models based on its precision (since it's a imbalanced dataset, accuracy is not the best metric to monitor).

Hyperparameters

180 Combinations of models to identify the best hyperparameters.

Model Building

Using the best model and best hyperparameters, generating a confusion matrix.

Predicting

Dropping features on the original dataset and predicting failures using our pipeline.

Conclusion

In conclusion, this project demonstrates the feasibility of using machine learning classification models for predicting machines failure. The best performing model can be used by companies to implement proactive maintenance strategies and reduce the costs associated with machine failure.