Final project in Prof. Louzoun's Unsupervised Learning course.
Two data sets were analyzed using five unsupervised learning methods. The first data set is of different types of hand postures. The second one describes a sample of pulsar candidates collected during the High Time Resolution Universe Survey (South). For each data set, the goal was to detect anomalous samples, cluster the data, visualize the clustering results, compute how well each clustering method fits the external classification, determine which clustering algorithm is better and explain the reason for the difference between them. Out of the five algorithms tested, according to the unsupervised methods, Spectral clustering with 2 clusters was the best algorithm for the data of hand postures. However, for the pulsar candidates data, K Means with 4 clusters provided the best results.
The data are too large to upload. They can be found here:
In order to run the code, the data sets shall be downloaded and placed in a directory named 'dataset'.
The main modules used on this project are:
- Sklearn
- Matplotlib
- Skfuzzy
- Numpy
- Pandas
- Scipy
- Yellowbrick
- torch