For two months, from May 27 2010 to July 27 2010, I recorded the number of visitors of this blog. I then fed this data into the "Statistics" node of the KNIME data mining tool.
Here are the results.
The minimum number of visitors per day is zero on Saturday May 29. The maximum number of visitors per day is 33 on Monday July 12, which is the date when I posted the statistics and fit measure for the Soccer World Cup 2010.
The average number of visitors per day is 12.387. I must say it is unbelievable that 12 people in average read my blog, especially considering the irregularity of my posting deadlines. The standard deviation is 7.376. This means that every day the number of visitors falls between 5 and 19.
The number of visitors is higher from Monday through Thursday and decreases sensibly on Friday and especially Saturday and Sunday. This means that most of you read my blog during their business hours :-)
Thus I decided to train a Bayesian Network to differentiate between weekend and not weekend days based on the number of visitors. I mapped the weekday into a binary variable (weekend/not weekend) and removed it from the original data. The Bayesian Predictor was then applied to the training data itself. I know it is not orthodox, but I just wanted to play with the KNIME Baysian classifier and I did not have much data at hand.
The Scorer node gave me an accuracy of 0.742. The confusion matrix tells me that 6 weekend days (out of 18) have been classified as not weekend and 10 not weekend days (out of 44) as weekend. I used the "Interactive Table" node to isolate such days and I saw that the mistaken weekend days were following some posting. The number of visitors was then higher than the average number of weekend visitors. Again the "Interactive Table" node told me that those days were far away from posting times, like yesterday, and the number of visitors was then exceptionally low.
The whole work was only an exercise to test KNIME for the implementation of:
- Bayesian Network (Learner and Predictor node)
- Interactive brushing ("Interactive Table" node)
- accuracy measures ("Scorer" node)
As you can see I have learned a lot of things about the average reader's weekly habits.
Thanks for following my blog. I will try to post more often to keep your interest and number of visits high.
Here are the results.
The minimum number of visitors per day is zero on Saturday May 29. The maximum number of visitors per day is 33 on Monday July 12, which is the date when I posted the statistics and fit measure for the Soccer World Cup 2010.
The average number of visitors per day is 12.387. I must say it is unbelievable that 12 people in average read my blog, especially considering the irregularity of my posting deadlines. The standard deviation is 7.376. This means that every day the number of visitors falls between 5 and 19.
The number of visitors is higher from Monday through Thursday and decreases sensibly on Friday and especially Saturday and Sunday. This means that most of you read my blog during their business hours :-)
Thus I decided to train a Bayesian Network to differentiate between weekend and not weekend days based on the number of visitors. I mapped the weekday into a binary variable (weekend/not weekend) and removed it from the original data. The Bayesian Predictor was then applied to the training data itself. I know it is not orthodox, but I just wanted to play with the KNIME Baysian classifier and I did not have much data at hand.
The Scorer node gave me an accuracy of 0.742. The confusion matrix tells me that 6 weekend days (out of 18) have been classified as not weekend and 10 not weekend days (out of 44) as weekend. I used the "Interactive Table" node to isolate such days and I saw that the mistaken weekend days were following some posting. The number of visitors was then higher than the average number of weekend visitors. Again the "Interactive Table" node told me that those days were far away from posting times, like yesterday, and the number of visitors was then exceptionally low.
The whole work was only an exercise to test KNIME for the implementation of:
- Bayesian Network (Learner and Predictor node)
- Interactive brushing ("Interactive Table" node)
- accuracy measures ("Scorer" node)
As you can see I have learned a lot of things about the average reader's weekly habits.
Thanks for following my blog. I will try to post more often to keep your interest and number of visits high.
RSS Feed
