Developed multiple ML Classifiers using SVM, PCA, Decision Tree, Random Forest, GMM & MLP Models for detecting Faults in Power Systems and comparing their accuracies & performances.
- The voltages and currents of all three phases were entered into a variety of machine learning models, after which the models were fine-tuned and their respective performances were compared.
- Using an SVM with a ‘rbf’ kernel, a Decision Tree with a maximum depth of 14, and a Random Forest with 162 estimators, the model was finetuned and achieved the highest accuracy possible (99.8%).
- Applied PCA to the data set and got more than 95% of the variance with 50% fewer input axes.
- Performed outlier detection with Gaussian Mixture Models (GMM) & obtained 99.5% accuracy at segregating faulty & non-fault conditions.
- Trained MLP models using a variety of activation functions and optimizers, and attained the following levels of accuracy: 73% when using linear and SGD, 98% when using relu and Adam, and 99.25% when using LeakyRelU and RMSProp.
- SVM-Decision_Tree-Random_Forest.ipynb - Training of SVM, Decision Tree & Random Forest & testing their accuracies.
- PCA-Random_Forest.ipynb - Reducing dimensions of dataset via PCA & achiving almost the same accuracy using Random Forest.
- Gaussian_Mixture_Model.ipynb - Training a Gaussian Mixture Models (GMM) & a performing Grid Search on no. of components & density threshold, to find the best parameters for the GMM accuracy.
- MLP_Models_Comparision.ipynb - Training 4 different MLP models & comparing thier accuracies & performances.