Supervised learning is straightforward on medical data using Orange, but unsupervised learning is more challenging. I selected a data set based on Florida County Health Ratings where individuals rated their health based on about 40 variables, from 67 counties. (I left the county names out). The data is continuous which is good. I added the preprocess widget so I could impute missing data and normalize the data. I used K-means clustering and with 2 clusters (C1,C2) the silhouette score was 0.223. I then connected the silhouette plot widget to a scatter plot and a data table. In the data table, most of the silhouette scores were 0.5-0.6. The scatter plot showed fairly good separation. On the X axis is the percent_fair/poor health rating and on the Y axis is physically unhealthy days per week. Clearly, it is separating individuals into healthy and unhealthy clusters. I opted to use the save data widget and export the clusters to a CSV file. I have done simple averages of other variables like family income and obesity and once again, C1 reflects healthy individuals and C2 reflects unhealthy individuals. I simply want to know is that how you would recommend analyzing the clusters? I also realize that I can convert the output of the two clusters into a classification model and evaluate the clusters that way. I will cut and paste my workflow and my scatterplot below. Thanks.

enter image description here

enter image description here


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.