Weka for clustering and regression itb vgsom
- 1. USING WEKA TO CLUSTERING AND REGRESSION ANALYSIS ( ITB PAPER ) ANURADHA CHAKRABORTY ROLL NO: 10BM60014 VINOD GUPTA SCHOOL OF MANAGEMENT, IIT KHARAGPUR
- 2. WEKA (Waikato Environment for Knowledge Analysis) is a popular suite of machine learning software written in Java, developed at the University of Waikato, New Zealand. WEKA is free software available under the GNU General Public License. WEKA is a unique software compared to MS –EXCEL because it can be used to run multivariate regression without any hassles. It also gives output showing dependent variable equation and other statistical data. Weka is a collection of machine learning algorithms for data mining tasks. The algorithms can either be applied directly to a dataset or called from your own Java code. Weka contains tools for data pre-processing, classification, regression, clustering, association rules, and visualization. It is also well-suited for developing new machine learning schemes. The initial versions of WEKA used only Attribute Relationship File Format (ARFF) files, saved as *.arff. But newer versions provide an option for multiple versions like: XRFF, Binary serial files, LIBSVM, SVM Light, CSV, C4.5 among others. USING WEKA: The WEKA GUI Chooser has the four following options: 1. Weka Explorer 2. Weka Experimenter 3. Weka Knowledge Flow 4. Simple CLI Weka Explorer has the following options in each tabs: 1. Preprocess 2. Classify 3. Cluster 4. Associate 5. Select Attributes 6. Visualize
- 3. Apart from doing these statistical operations, each of the data can be visualized graphically and filtered according to requirement. Weka Experimenter: There are several algorithms for each process. Thus the criticality of the software lies in identifying the optimal algorithm. For Regression and classification, Experimenter gives a comparisn of the best algorithm by statistical analysis. Unfortunately, such an option is not there for Clustering algorithms. Import of data: Data is imported in form of CSV file which is converted into arff format automatically while importing. The data is imported through Preprocess tab of WEKA as shown in picture above. CLUSTERING Definition: Cluster analysis is a class of statistical techniques that can be applied to data that exhibit “natural” groupings. Cluster analysis sorts through the raw data and groups them into clusters. A cluster is a group of relatively homogeneous cases or observations. Objects in a cluster are similar to each other. They are also dissimilar to objects outside the cluster, particularly objects in other clusters.”
- 4. DATA SET USED FOR CLUSTERING The example used is a survey report on instant noodles. It had: Instances: 76 Attribute: 33 The questions or attributes were as follows: Age Profession Diabetesstop Obesitystop Otherstop Cadburynchocl Homemadesweets Sweetfrmshop Cakepastry Sugarcube Celebration Gifts Beginningauspicious Yummyfood Healthconcern Lunchdinnerafter Tastytraditn Abroad Frequencyeating Inflnearby Inflfrndrelative Inflblogonline Advert Quality Packaging Ambience Price Imptraditonsweet Newexperimentswt Newvariety Homedeliveryimp Impchitchatplace Packagdsweetslngtime
- 5. PROCEDURE AND RESULT: Data-set is taken from my AMRP project survey, regarding the interest and motivation of consumers towards traditional sweets. Simple K-Mean Algorithm was used to cluster the data set. The output is as follows: Attribute Full Data 0 1 (76) (44) (32) ======================================================= Age 1.6711 1.6364 1.7188 Profession 1.7632 1.6818 1.875 Diabetesstop 2.3553 2.3636 2.3438 Obesitystop 1.9605 1.9545 1.9688 otherstop 1.9474 1.8636 2.0625 Cadburynchocl 4.2895 4.25 4.3438 homemadesweets 4.3421 4.3636 4.3125 sweetfrmshop 4.0395 4.1136 3.9375 cakepastry 3.9342 4.0455 3.7813 sugarcube 2.4605 2.5 2.4063 celebration 4.1447 4.3409 3.875 gifts 3.7632 3.7955 3.7188 beginningauspicious 3.7763 3.8636 3.6563 yummyfood 3.8158 3.9318 3.6563 healthconcern 2.9868 3 2.9688 lunchdinnerafter 3.9737 4.0909 3.8125 tastytraditn 3.7632 4.0227 3.4063 abroad 1.8684 1.8864 1.8438 frequencyeating 2.5658 2.4318 2.75 inflnearby 3.0 4.0 3.0 inflfrndrelative 4.0 4.0 3.0 inflblogonline 3.0 3.0 2.0 advert 3.0 3.0 2.0 quality 5.0 5.0 5.0 packaging 3.0 3.0 4.0 ambience 3.0 3.0 4.0 price 3.0 4.0 3.0 imptraditonsweet 5.0 5.0 3.0 newexperimentswt 3.0 3.0 3.0 newvariety 3.0 3.0 4.0 homedeliveryimp 2.8158 2.8409 2.7813 impchitchatplace 3.3421 3.3182 3.375
- 6. packagdsweetslngtime 3.1579 3 3.375 Note: The significant values in the above table, on which the cluster characteristics are formed, are marked with red. Clustered Instances 0 44 ( 58%) 1 32 ( 42%) INTERPRETATION: ASPECTS CLUSTER ‘0’ CLUSTER ‘1’ Traditionality Loves traditional sweets. Loves experiments and newer Considers sweet as a variety of sweets traditional symbol. Wants sweet after lunch or dinner. Frequency of consumption High Medium Price More price sensitive Lesser price sensitive. Influnce by friends and High Medium. Generally tries new relatives or advertisements to shop by own instinct. try a new shop Ambience of shop and Matters less Matters significantly. packaging Food Court for chatting (Like preferred prefered Haldiram) Packaged/ tinned sweets Medium Good Demand INFERENCE AND SUGGESTION DERIVED FROM THE CLUSTERING: There are two distinct clusters of consumers in the sweet industry. Cluster ‘0’ (58%) considers sweet as the “symbol of tradition”, which is typically savored after lunch and dinner. They enjoy the most traditional sweets, and don’t prefer to try new variants. They prefer sticking to old shops unless inspired by external agents (friends/ relatives/ blog/ advertisements etc) to try otherwise. Quality is an important factor. But ambience and packaging doesn’t play a major role. So, shops like Nokur or Girish Dey will be their typical favorite ones. Cluster ‘1’(42%) are the true connoisseurs of sweets. They appreciate both traditional as well as experimental sweets (the new variants). They often prefer trying out new shops and brands. Packaged sweets are also preferred which can be savored later. Apart from quality, ambience and packaging plays a vital role, where as price is of medium importance. This
- 7. cluster seems to be more impulsive consumers, and would probably not mind paying a premium for some new and creative sweets. So, brands like K.C. Das will be their preferred choice. REGRESSION The next procedure is regression analysis. We obtain data from stores on monthly sales of a celebration chocolate pack depending on the amount spent on its promotion in terms of posters used around the block or any other effort . Here after we select all attributes and go to classify tab and run regression function. OUTPUT The output obtained is given below = Run information === Scheme: weka.classifiers.functions.LinearRegression -S 0 -R 1.0E-8 Relation: Problem_2-weka.filters.unsupervised.attribute.Remove-R1 Instances: 46
- 8. Attributes: 3 Sales Price Promotion Test mode: split 80.0% train, remainder test === Classifier model (full training set) === Linear Regression Model Sales = -53.2173 * Price + 3.6131 * Promotion + 5837.5208 Time taken to build model: 0 seconds === Evaluation on test split === === Summary === Correlation coefficient 0.8066 Mean absolute error 543.6332 Root mean squared error 711.4575 Relative absolute error 48.288 % Root relative squared error 59.6886 % Total Number of Instances 5 Ignored Class Unknown Instances 4 INTERPRETETION The given data shows correlation coefficient of 0.8066 which means 65% accuracy of the model. As expected we find that sales will decrease due to increase in price and increase with increase in promotion budget. This explains how WEKA can be used for multivariate regression . REFERENCE http://en.wikipedia.org/wiki/Weka_(machine_learning) http://www.cs.waikato.ac.nz/ml/weka/ http://en.wikipedia.org/wiki/Cluster_analysis_(in_marketing)