Description: Implementation of classification algorithms in 2020 for the elective course "Pattern Recognition" for my undergraduate studies in Computer Science and
Engineering at University of Ioannina. The two data experiments named spambase.data (source) and credit card clients dataset.xls (source) are problems of binary classification.
These experiments will be utilized for the needs of this assignment.
Programming language: Python
Interpreter: Python 3.6
Software used: PyCharm (vers2019.3.5)
Classification methods that are being implemented:
-
LVQ algorithm with the specific variation of dynamically spliting the space of the data into spherical Gaussian areas of the same category. The process of the classified decision for a foreign element will be determined by the category of the nearest Gaussian area.
-
Nearest Neighbor k-NN algorithm using the Euclidean distance (variable k must be determined by the user each time).
-
Neural Network with Sigmoid as the activation function consisted of a) 1 hidden layer and different number K of neurons and b) 2 hidden layers and different number K1 and K2 neurons per layer.
-
Support Vector Machines (SVM) using a) linear kernel and b) gaussian kernel.
-
Naive Bayes classifier with normal distribution for each element.
Each of the above classification methods must be evaluated with the following evaluation metrics:
-
a) Accuracy:
Accuracy = (TP+TN)/(P+N)b) F1 score:
F1score = 2*(Precision*Recall)/(Precision+Recall) Precision = TP/(TP+FP) Recall = TP/(TP+FN)
where: TP: True Positives, TN: True Negatives, FP: False Positives, FN: False Negatives, P: Positives, N: Negatives
- 10-folds cross validation: specifically, the original data will be randomly splited into 10 subsets and for each of these subsets the performance of the method (testing set) will be determined by training the rest 9 subsets (training set) with it.
Greek presentation of the assignment at PatternRecognition2020_Homework1.pdf file.