Author

Name: Wenjun Wu

GT username: wwu94

Project Info

Implementation Platform: Weka 3.8.1 with Student Filters extension (https://github.com/cgearhart/students-filters.)

Wilt Dataset:

~/wwu94/Silt/wilt.arff

Car Evaluation Dataset:

~/wwu94/Car/car.arff

Clustering Analysis

K-means Clustering (KM)

​ Weka —> Clustering —> SimpleKMeans

​ MaxIteration = 10000

Expectation Maximization Clustering

​ Weka —> EM

​ MaxIteration = 10000

`

Dimensionality Reduction

ICA

​ Weka --> Preprocess-->Unsupervised --> attribute--> IndependenComponents

​ Kurtosis were calculated using python kurtosis_calculator *.arff

PCA

​ Weka --> Preprocess-->Unsupervised --> attribute--> PrincipalComponents

​ Eigenvalues were generated using SelectAttribute —> PrincipalComponents

RP

Weka --> Preprocess-->Unsupervised --> attribute--> RandomProjection

Information Gain Feature Selection

Weka --> Preprocess-->supervised --> attribute--> FeatureSelection-->Information Gain (rankers)

Neural Network Learning

1. Load the dataset and then preprocess the raw dataset or load the preprocess dataset in ~/wwu94/Wilt/preprocess
2. Go to 'Classify' interface. Choose the MultilayerPerceptron. Use Percentage split with 70% training set.Run the neural network algorithm with parameters described in the report.

Clustering as feature reduction + NN learning

1. Load the preprocess dataset in ~/wwu94/Wilt/preprocess
2. Use AddCluster Filter and select EM or SimpleKMeans. Change parameters as described in the report.
3. Click 'Apply' and delete all attributes except 'Class' and 'Cluster'.
4. Go to 'Classify' interface. Choose the MultilayerPerceptron. Use Percentage split with 70% training set. Run the neural network algorithm with parameters described in the report.