Features

Recent KDnuggets Poll asked: What data mining techniques you use regularly? This poll was a multiple choice, and an average voter used 3.6 methods.

Based on 768 choices from 213 voters, the top techniques were:

• Decision Trees/Rules, 16%
• Clustering, 12%
• Statistics, 12%

While this poll should not be considered scientific, it is interesting to compare results with an earlier poll on Data Mining Techniques (Oct 2002). The most remarkable is that most methods (including Decision Trees, Clustering, Logistic Regression, Nearest Neighbor) kept almost the same share of the usage.

In the following table, we compare Use Share computed as percentage of voters who used a particular techniques (i.e. the count of votes divided by the total number of voters, 212 in 2002 and 213 in 2003), rather than count of votes divided by the total number of votes used above.

Method	2002 votes	Use Share	2003 votes	Use Share	Change
Decision Trees/Rules	128	60.1%	120	56.6%	-5.8%
Clustering	103	48.4%	93	43.9%	-9.3%
Statistics	101	47.4%	92	43.4%	-8.5%
Neural networks	75	35.2%	71	33.5%	-4.9%
Logistic regression	75	35.2%	69	32.5%	-7.6%
Visualization	52	24.4%	55	25.9%	6.3%
Association rules	63	29.6%	42	19.8%	-33.0%
Nearest neighbor	42	19.7%	38	17.9%	-9.1%
Text mining	30	14.1%	30	14.2%	0.5%
Web mining	19	8.9%	29	13.7%	53.4%
SVM			24	11.3%
Bayesian nets / Naive Bayes	24	11.3%	24	11.3%	0.5%
Sequence analysis	27	12.7%	24	11.3%	-10.7%
Hybrid methods	21	9.9%	23	10.8%	10.0%
Genetic algorithms	26	12.2%	12	5.7%	-53.6%
Other	20	9.4%	22	10.4%	10.5%
Voters	212		213		0

The biggest gainers were:

• SVM, with 11% usage in 2003 (not asked in 2002 poll)
• Web mining, 53% increase

• Genetic algorithms, 53% decline
• Association Rules, 33% decline