Unsupervised Machine Learning
Process and analyze a cryptocurrency dataset to fit the unsupervised machine learning model, create a report that includes what cryptocurrencies are on the trading market and how they could be grouped to create a classification system for a new investment in cryptocurrency.
-
Preprocessing the data por PCA
-
Reducing Data Dimensions Using PCA
-
Clustering Cryptocurrencies Using K-means
-
Visualizing Cryptocurrencies Results
The dataframe above shows the three principal components (PCAs) generated, where data is aggregated together by the similarity of the features.
Defining the k value on a range from 1 to 10, then using the K-mean algorithm to iterate through the k value, find their inertia and store it into a dataframe, that can then be used to generate the elbow curve plot shown below.
After analyzing the elbow curve, it can be seen that the angle at point 4 looks like an elbow, which means using 4 clusters for the analysis is a good option.
The above 3D Plot takes in 3 axis; x, y, and z corresponding the three PCAs previously calculated. This plot confirms that dividing the groups into four cluster makes sense, where we can see that there are four distinct groups that correspond to the four clusters that we expected the model to break the data into.
The scatter plot above is generated without the PCA values and that is why it is harder to see the four distinct clusters, in other words since we cannot see the depth of the points it is hard to tell if they were ideally grouped.
The best way to see the results was represented by the 3D scatter plot that takes in the PCA values as 3 axis, where four different clusters can be seen. The clusters represented by colors yellow and purple are the ones with more concentration of points. Something that has caught my attention was the class 1, represented by the orange X point on the 3D plot, the fact that it is a single point and is very far from all the other points could mean that it is an outlier, which could be skewing our data into four clusters, which could have easily been only three. I suggest that this specific point should be investigated, possibly removed, and the results recalculated. In addition to that the other clusters should be analized using the dataframes for each class to check for patterns and relations that made them to be grouped together.