Laser-Detection-Machine-Learning

About The Project

This machine learning project is a part of my Master's course Machine Learning to classify faulty and non-faulty lasers correctly.

Problem Statement

For the purpose of quality assurance, a manufacturer of medical lasers wants to introduce a system which recognizes defective products. For the lasers produced, a constant lightoutput with a frequency as constant as possible is desired. Certain fluctuations are acceptable; Lasers in which the power fluctuates to an intolerable extent should be sorted out.For this purpose, the intensity of each laser is measured for one minute – one measurement per second.

Implementation

• four classifiers: Decision Tree, Random Forest, SVM(Support Vector Machine), KNN (K Nearest Neighbor)
• evaluation/permormance metrics: accuracy, recall, precision, f1-score, confusion matrix, AUC, ROC curve
• dataset: laser.mat

Justify model's robustness

Following this steps to justify all model's robustness:

• Model training using default parameters
• Generating confusion matrix, classification report and ROC curve
• Applying GridSearchCv for finding best parameters
• Model training using best parameters
• Generating confusion matrix, classification report and ROC curve for comparing with the default parameters

Conclusion

Among the 4 models, we did not get any improvement after hyper-parameter tuning on the two following models:

• SVM (98%)
• K-Nearest Neighbor (95%)

On the other hand, Following two models showed an improvement after hyper-parameter tuning:

• Decision Tree (from 80% to 90%)
• Random Forest (from 96% to 98%)

Therefore, Random Forest is outperforming all other models.

Credit Card Default Prediction

(back to top)

Given data about credit card clients, we try to predict whether a given client will default or not.

Logistic regression, support vector machine, and neural network models used to make our predictions.

Data source: https://www.kaggle.com/datasets/uciml/default-of-credit-card-clients-dataset