Machine-Learning-Data-Clustering

Java machine learning to perform data clustering. Compare the performance of three clustering algorithms using the Iris flower dataset.

Algorithms

• KMeans
• Density Based Spatial Clustering
• KMediods

Performance Evaluation

AICScore:

This method evaluates the overall fitness of the model with the data set. It weighs on parsimony, that simpler models are more effective than more complex models. It evaluates the models based on the fitness, how well the data fits with the model, and the complexity, how many parameters are used to describe the model. This evaluation method was chosen because it gives an insight as to which model is the best fit for the dataset by giving a lower score for the model that had the best balance between complicity and accuracy.

Hybrid Pairwise Similarities:

This evaluation method is used to evaluate the similarity in pairs of data points in a cluster. This evaluation method was used because it uses multiple metrics such as euclidean distance, cosine similarity, and correlation coefficient to benchmark the relationship between different data points.

GPlus:

This evaluation method compares the similarity of the objects in a cluster to the difference in other clusters. This is why it is initialized with Euclidean distance since it is used to calculate the distance in the clustering models. The reason this evaluation method was chosen was because it designed the handle datasets with a known number of classes.

Evaluations

KMeans:

• AICScore: 9938.781284138158
• Hybrid Pairwise Similarities: 1.019093657866949
• GPlus: 4.385380090967006E-5

Density Based Spatial Clustering:

• AICScore: 9708.693384443968
• Hybrid Pairwise Similarities: 1.018305789497345
• GPlus: 6.638665117644806E-5

KMedoids:

• AICScore: 9927.020769913317
• Hybrid Pairwise Similarities: 1.0194096168845206
• GPlus: 4.431828601791711E-5