rubenandrebarreiro/banknote-authentication-classification

💵 🧠 📈 A project based in Machine Learning, in the topic of Supervised Learning. This project was built using Python, NumPy, SciKit-Learn, Anaconda and Spyder. The scenario of the project was a parameterization, fitting and comparison of the Logistic Regression, Naïve Bayes with custom KDEs (Kernel Density Estimations) and Gaussian Naïve Bayes Classifiers. The dataset is inspired on the Banknote Authentication problem in the UCI (University of California, Irvine) Machine Learning repository. The final goal of the project was to implement and tune the Classifiers, by computing some metrics in the samples of the Training Set (together with the Validation Set), such as, the Training Errors and Validation Errors, varying some Regularization Parameters and Hyperparameters, in order to reduce those Errors, to getting the best models, plotting also, those Training Errors and Validation Errors. After the Classifiers be tuned, the Classifiers try to predict the more accurately possible the samples in the Testing Set, computing also their True/Testing Errors. It was computed also some Comparison and Statistical Test Methods, such as, the Approximate Normal Test and the McNemar Test, in order to give a more specific comparison between the Classifiers.

PythonMIT