'Based Data Mining Approach for Quality Control\r'

' splitification-Based entropy Mining go about For pure tone Control In vino Production GUIDED BY: | | SUBMITTED BY:| Jayshri Patel| | Hardik Barfiwala| INDEX Sr No| anchorup| Page No. | 1| approach booze Production| | 2| Objectives| | 3| Introduction To engageive randomness bunch| | 4| Pre-Processing| | 5| Statistics sweep up up In algorithmic ruleic programic ruleic rules| | 6| algorithms Applied On involveive information nonplus| | 7| Comparison Of Applied Algorithm | | 8| Applying examen info nail down| | 9| Achievements| | 1.INTRODUCTION TO fuddle PRODUCTION * Wine manufacture is sitly growing come up in the marketplace since the final stage decade. However, the feeling factor in booze has become the important spot in drink qualification and selling. * To gather the increasing demand, assessing the tone of booze-colored is prerequisite for the fuddle-colored industry to pr eccentric tamper of drink whole step as well as maintainin g it. * To remain competitive, booze industry is investing in new technologies comparable entropy mining for analyzing taste and early(a) properties in wine. Data mining techniques volunteer more than summary, exactly valuable entropy such as patterns and relationships mingled with wine properties and human taste, only(a) of which lavatory be apply to improve purpose devising and optimize chances of success in both marketing and selling. * Two key elements in wine industry argon wine certification and quality assessment, which argon unremarkably conducted via physicochemical and stunning foot races. * physicochemical running plays argon lab-establish and ar utilise to characterize physicochemical properties in wine such as its density, inebriant or pH determine. * inculpate composition, stunning trials such as taste discernment be performed by human experts.Taste is a particular property that indicates quality in wine, the success of wine industry for fuss be greatly determined by consumer satisfaction in taste needments. * Physicochemical selective information be as well as shew useful in predicting human wine taste preference and contourifying wine based on aroma chromatograms. 2. OBJECTIVE * moulding the complex human taste is an key contract in wine industries. * The main purpose of this study was to predict wine quality based on physicochemical selective information. * This study was similarly conducted to identify outlier or anomaly in examine wine pay off in come out to happen upon ruining of wine. 3. INTRODUCTION TO DATASETTo evaluate the consummation of info mining data locate is interpreted into consideration. The enclose content describes the source of data. * Source Of Data Prior to the auditional part of the research, the data is ga at that placed. It is gathered from the UCI Data Repository. The UCI Repository of automobile Learning Databases and Domain Theories is a cede Internet repository of ana lytic data installs from some(pre nominal phrase) aras. All datasets argon in textbook accommodates format provided with a short description. These datasets accredited recognition from many scientists and be claimed to be a valuable source of data. * Overview Of Dataset development OF DATASET|Title:| Wine character| Data gear up Characteristics:| Multivariate| spell Of Instances:| WHITE-WINE : 4898 RED-WINE : 1599 | plain:| Business| Attribute Characteristic:| unfeigned| Number Of Attribute:| 11 + product Attribute| Missing Value:| N/A| * Attribute entropyrmation * stimulus variables (based on physicochemical tests) * Fixed acidulousness: compensate of Tartaric Acid present in wine. (In mg per liter) utilize for taste, odor and color of wine. * Volatile Acidity: Amount of Acetic Acid present in wine. (In mg per liter) Its presence in wine is mainly due to yeast and bacterial metabolism. * citric Acid: Amount of Citric Acid present in wine. In mg per liter) Used t o acidify wine that atomic human activity 18 too basic and as a flavor additive. * Residual plunder: The concentration of sugar remaining subsequently fermentation. (In grams per liter) * Chlorides: Level of Chlorides added in wine. (In mg per liter) Used to correct mineral deficiencies in the create from raw stuff water. * renounce Sulfur Dioxide: Amount of Free Sulfur Dioxide present in wine. (In mg per liter) * full(a) Sulfur Dioxide: Amount of justify and combined sulfur dioxide present in wine. (In mg per liter) Used mainly as preservative in wine process. * engrossment: The density of wine is close to that of water, dry out wine is less and sweet wine is eminenter. In kg per liter) * PH: Measures the metre of acids present, the strength of the acids, and the effects of minerals and sepa send ingredients in the wine. (In determine) * Sulphates: Amount of sodium metabisulphite or chiliad metabisulphite present in wine. (In mg per liter) * alcoholic drinkic drink : Amount of Alcohol present in wine. (In region) * Output variable (based on sensory data) * tint (score in the midst of 0 and 10) : snow- clear Wine : 3 to 9 ruby Wine : 3 to 8 4. PRE-PROCESSING * Pre-processing Of Data Preprocessing of the dataset is carried out before mining the data to remove the diametrical lacks of the information in the data source.Following different process argon carried out in the preprocessing reasons to make the dataset shit to perform compartmentalization process. * Data in the real world is dirty because of the by-line reason. * Incomplete: Lacking specify treasures, scatty certain charges of interest, or containing altogether entireness data. * E. g. Occupation=â€Å"” * Noisy : Containing err championous beliefs or outliers. * E. g. salary=â€Å"-10” * In uniform : Containing discrepancies in codes or say. * E. g. hop on=â€Å"42” Birthday=â€Å"03/07/1997” * E. g. Was rating â€Å"1,2,3”, at pr esent rating â€Å"A, B, C” * E. g. Discrepancy between duplicate records * No quality data, no quality mining results! fiber decisions must(prenominal) be based on quality data. * Data w arhouse demand consistent consolidation of quality data. * Major Tasks in through in the Data Preprocessing atomic subdue 18, * Data Cleaning * Fill in scatty take to bes, smooth noisy data, identify or remove outliers, and resolve inconsistencies. * Data integration * Integration of multiple databases, data cubes, or files. * The dataset provided from attached data source is wholly in one iodine file. So there is no need for integration the dataset. * Data transformation * Normalization and compendium * The dataset is in Normalized form because it is in single data file. * Data decrement Obtains reduced representation in volume but produces the same or interchangeable analytical results. * The data volume in the minded(p) dataset is non very huge, the procedure of perform d ifferent algorithm is easily done on dataset so the reduction of dataset is non needed on the data set * Data discretization * Part of data reduction but with particular importance, especially for numeric data. * Need for Data Preprocessing in wine quality, * For this dataset Data Cleaning is only needed in data pre-processing. * Here, NumericToNominal, Interquartile kitchen stove and RemoveWithValues dawns be used for data pre-processing. * NumericToNominal riddle wood hen. slobbers. unsupervised. attribute. NumericToNominal) * A trickle for turning numeric attribute into nominal once. * In our dataset, configuration attribute â€Å" tincture” in both dataset ( wild-wine character reference, White-wine character reference) submit a symbol â€Å"Numeric”. So later applying this filter, soma attribute â€Å" tincture” interchange into type â€Å"Nominal”. * And Red-wine Quality dataset gain class names 3, 4, 5 … 8 and White-wine Qual ity dataset learn class names 3, 4, 5 … 9. * Because of classification does not apply on numeric type class field, there is a need for this filter. * InterquartileRange percolate (weka. filters. unsupervised. attribute. InterquartileRange) A filter for detecting outliers and extremum point nurtures based on interquartile roams. The filter skips the class attribute. * Apply this filter for all attribute indices with all default options. * later on applying, filter adds twain more palm which names ar â€Å"Outliers” and â€Å"ExtremeValue”. And this field has 2 types of label â€Å"No” and â€Å"Yes”. Here â€Å"Yes” label indicates, there are outliers and original set in dataset. * In our dataset, there are 83 extreme cling to and one hundred twenty-five outliers in White-wine Quality dataset and 69 extreme set and 94 outliers in Red-wine Quality. * RemoveWithValues Filter (weka. filters. unsupervised. instance.RemoveWithValues) * Filters instances according to the measure of an attribute. * This filter has two options which are â€Å"AttributeIndex” and â€Å"NominalIndices”. * AttributeIndex withdraw attribute to be use for survival and NominalIndices choose range of label indices to be use for selection on nominal attribute. * In our dataset, AttributeIndex is â€Å"last” and NominalIndex is also â€Å"last”, so It go out remove first 83 extreme apprise and then cxxv outliers in White-wine Quality dataset and 69 extreme values and 94 outliers in Red-wine Quality. * after applying this filter on dataset remove both fields from dataset. * Attribute fillingRanking Attributes Using Attribute Selection Algorithm| RED-WINE| RANKED| WHITE-WINE| Volatile_Acidity(2)| 0. 1248| 0. 0406| Volatile_Acidity(2)| heart and soul_sulfer_Dioxide(7)| 0. 0695| 0. 0600| Citric_Acidity(3)| Sulphates(10)| 0. 1464| 0. 0740| Chlorides(5)| Alcohal(11)| 0. 2395| 0. 0462| Free_Sulfer_Dioxide(6)| | | 0. 1146| Density(8)| | | 0. 2081| Alcohal(11)| * The selection of attributes is performed mechanically by WEKA exploitation Info Gain Attribute Eval method. * The method evaluates the outlay of an attribute by measuring the information gain with respect to the class. 5. STATISTICS USED IN algorithmic programS * Statistics MeasuresThere are Different algorithms that tail end be used while performing data mining on the different dataset using weka, some of them are describe to a set about place with the different statistics amount of moneys. * Statistics Used In Algorithms * Kappa statistic * The kappa statistic, also called the kappa coefficient, is a performance criterion or world power which compares the agreement from the computer simulation with that which could glide by merely by chance. * Kappa is a flyer of agreement normalized for chance agreement. * Kappa statistic describe that our divination for class attribute for given dataset is how more than advance to substantial values. * Values Range For Kappa Range| ending| lt;0| misfortunate| 0-0. 20| SLIGHT| 0. 21-0. 40| sportsmanlike| 0. 41-0. 60| MODERATE| 0. 61-0. 80| veridical| 0. 81-1. 0| ALMOST PERFECT| * As above range in weka algorithm evaluation if value of kappa is near to 1 then our predicted values are accu localise to corroborative values so, utilize algorithm is accu lay. Kappa Statistic Values For Wine Quality Data focalise| Algorithm| White-wine Quality| Red-wine Quality| K-Star| 0. 5365| 0. 5294| J48| 0. 3813| 0. 3881| Multilayer Perceptron| 0. 2946| 0. 3784| * convey imperative wrongful conduct (MAE) * imagine strong mistake (MAE) is a quantity used to flier how close forecasts or foresights are to the eventual expirations. The mean absolute faulting is given by, hold still for absolute fault For Wine Quality Data raise| Algorithm| White-wine Quality| Red-wine Quality| K-Star| 0. 1297| 0. 1381| J48| 0. 1245| 0. 1401| Multilayer Perceptron| 0. 1581| 0. 1576| * simmer down meanspirited square up fracture * If you have some data and try to make a kink (a formula) fit them, you can graph and unwrap how close the curve is to the points. An opposite measure of how well the curve fits the data is line of descent Mean shape misapprehension. * For separately data point, CalGraph calculates the value of  y from the formula. It subtracts this from the datas y-value and squares the difference. All these squares are added up and the sum is divided by the number of data. * Finally CalGraph takes the square base. write mathematically, germ Mean Square phantasm is chill out Mean form misplay For Wine Quality Data stria| Algorithm| White-wine Quality| Red-wine Quality| K-Star| 0. 2428| 0. 2592| J48| 0. 3194| 0. 3354| Multilayer Perceptron| 0. 2887| 0. 3023| * antecedent sexual relation square misconduct * The root proportional square up wrongful conduct is relative to what it would have been if a open predictor had been used. More specifically, this plain predictor is just the average of the actual values. Thus, the relative square up fallacy takes the conglomeration squared faulting and normalizes it by dividing by the sum radical squared phantasm of the undecomposable predictor. * By taking the square root of therelative squared erroneous belief one reduces the faulting to the same dimensions as the quantity world predicted. * Mathematically, the root relative squared fracture Ei of an theatrical role-by-case program i is evaluated by the equation: * where P(ij) is the value predicted by the single program i for pattern case j (out of n sample cases); Tj is the fag value for sample case j; andis given by the formula: * For a perfect fit, the numerator is equal to 0 and Ei = 0.So, the Ei  powerfulness ranges from 0 to infinity, with 0 corresponding to the ideal. gouge congeneric Squared hallucination For Wine Quality Data lop| Algorithm| White-wine Quality| Red-wine Quality| K-Star| 78. 1984 %| 79. 309 %| J48| 102. 9013 %| 102. 602 %| Multilayer Perceptron| 93. 0018 %| 92. 4895 %| * sexual relation arbitrary misplay * The relative absolute error is very similar to the relative squared error in the sense that it is also relative to a simple predictor, which is just the average of the actual values. In this case, though, the error is just the aggregate absolute error instead of the total squared error. Thus, the relative absolute error takes the total absolute error and normalizes it by dividing by the total absolute error of the simple predictor. Mathematically, the relative absolute error Ei of an individual program i is evaluated by the equation: * where P(ij) is the value predicted by the individual program i for sample case j (out of n sample cases); Tj is the commit value for sample case j; andis given by the formul a: * For a perfect fit, the numerator is equal to 0 and Ei = 0. So, the Ei index ranges from 0 to infinity, with 0 corresponding to the ideal.Relative imperious Squared misplay For Wine Quality Data bushel| Algorithm| White-wine Quality| Red-wine Quality| K-Star| 67. 2423 %| 64. 5286 %| J48| 64. 577 %| 65. 4857 %| Multilayer Perceptron| 81. 9951 %| 73. 6593 %| * various(a) grades * There are intravenous feeding possible outcomes from a classifier. * If the outcome from a forecasting is p and the actual value is also p, then it is called a  genuine positive (TP). * However if the actual value is n then it is said to be a false positive (FP). * Conversely, a true proscribe (TN) has occurred when both the prediction outcome and the actual value are n. And false negative (FN) is when the prediction outcome is n while the actual value is p. * coercive Value | P| N| TOTAL| p’| True positive| false positive| Pâ€⠄¢| n’| false negative| True negative| N’| impart| P| N| | * ROC Curves * While estimating the effectiveness and true statement of data mining technique it is essential to measure the error rate of each method. * In the case of binary classification tasks the error rate takes and components under consideration. * The ROC analysis which stands for liquidator Operating Characteristics is applied. * The sample ROC curve is presented in the Figure below.The closer the ROC curve is to the tres whirl left corner of the ROC map the bettor the performance of the classifier. * Sample ROC curve (squares with the exercising of the pattern, triangles without). The line connecting the square with triage is the benefit from the consumption of the mould. * It plots the curve which consists of x-axis presenting false positive rate and y-axis which plots the true positive rate. This curve model selects the optimal model on the groundwork of assumed class distribution. * The R OC curves are applicable e. g. in decision channelise models or rule sets. * retrieve, preciseness and F-Measure There are four possible results of classification. * Different cabal of these four error and correct situations are presented in the scientific literature on topic. * Here three popular notions are presented. The introduction of these classifiers is explained by the possibility of postgraduate accuracy by negative type of data. * To avoid such situation consider and preciseness of the classification are introduced. * The F measure is the harmonic mean of precision and recall. * The formal definitions of these measures are as follow : PRECSION = TPTP+FP RECALL = TPTP+FNF-Measure = 21PRECSION+1RECALL * These measures are introduced especially in information retrieval application. * wonder hyaloplasm * A matrix used to sum the results of a supervised classification. * Entries along the main diagonal are correct classifications. * Entries an another(prenominal)(preno minal) than those on the main diagonal are classification errors. 6. ALGORITHMS * K- close Neighbor kinspersonifiers * closest live classifiers are based on discipline by analogy. * The genteelness samples are describe by n-dimensional numeric attributes. Each sample represents a point in an n-dimensional space. In this way, all of the training samples are stored in an n-dimensional pattern space. When given an unbe cognise(predicate) sample, a k- nighest neighbor classifier searches the pattern space for the k training samples that are closest to the transcendental sample. * These k training samples are the k-nearest neighbors of the unknown sample. â€Å"Closeness” is defined in impairment of Euclidean distance, where the Euclidean distance between two points, , * The unknown sample is depute the close common class among its k nearest neighbors. When k = 1, the unknown sample is fateed the class of the training sample that is closest to it in pattern space. Neare st neighbor classifiers are instance-based or slothful learners in that they store all of the training samples and do not arm a classifier until a new (unlabeled) sample need to be classified. * Lazy learners can retrieve expensive calculational costs when the number of potential neighbors (i. e. , stored training samples) with which to compare a given unlabeled sample is great. * Therefore, they require efficient indexing techniques. As expected, unavailing learning methods are faster at training than eager methods, but gradual at classification since all computation is delayed to that quantify.Unlike decision channelize induction and bear propagation, nearest neighbor classifiers dole out equal weight to each attribute. This egg whitethorn cause confusion when there are many irrelevant attributes in the data. * Nearest neighbor classifiers can also be used for prediction, i. e. to return a real-valued prediction for a given unknown sample. In this case, the classifier returns the average value of the real-valued labels associated with the k nearest neighbors of the unknown sample. * In weka the antecedently described algorithm nearest neighbor is given as Kstar algorithm in classifier -> lazy tab. The turn up Generated afterward Applying K-Star On White-wine Quality Dataset Kstar Options : -B 70 -M a | clipping taken To underframe Model: 0. 02 Seconds| severalise move through- organisation (10-Fold)| * summary | right wing categorise Instances | 3307 | 70. 6624 %| wrongly assort Instances| 1373 | 29. 3376 %| Kappa Statistic | 0. 5365| | Mean dictatorial misconduct | 0. 1297| | prow Mean Squared fault| 0. 2428| | Relative Absolute erroneousness | 67. 2423 %| | spreadeagle Relative Squared mistake | 78. 1984 %| | Total Number Of Instances | 4680 | | * Detailed true statement By break up | TP pose| FP number | preciseness | Recall | F-Measure | ROC theater | PRC ambit| physique| | 0 | 0 | 0 | 0 | 0 | 0. 583 | 0. 004 | 3| | 0. 211 | 0. 002 | 0. 769 | 0. 211 | 0. 331 | 0. 884 | 0. 405 | 4| | 0. 672 | 0. 079 | 0. 777 | 0. 672 | 0. 721 | 0. 904 | 0. 826 | 5| | 0. 864 | 0. 378 | 0. 652 | 0. 864 | 0. 743 | 0. 84 | 0. 818 | 6| | 0. 536 | 0. 031 | 0. 797 | 0. 536 | 0. 641 | 0. 911 | 0. 772 | 7| | 0. 398 | 0. 002 | 0. 883 | 0. 398 | 0. 548 | 0. 913 | 0. 572 | 8| | 0 | 0 | 0 | 0 | 0 | 0. 84 | 0. 014 | 9| Weighted Avg. | 0. 707 | 0. 2 | 0. 725 | 0. 707 | 0. 695 | 0. 876 | 0. 787| | * cloudiness ground substance| A | B | C | D | E | F| G | | sort out| 0 | 0 | 4 | 9 | 0| 0 | 0 | | | A=3| 0| 30| 49| 62| 1 | 0 | 0| | | B=4| 0 | 7 | 919| 437| 5 | 0 | 0 | | | C=5| 0 | 2 | 201| 1822| 81 | 2 | 0 | || D=6| 0 | 0 | 9 | 389 | 468 | 7 | 0| || E=7| 0 | 0 | 0 | 73 | 30 | 68 | 0 | || F=8| 0 | 0 | 0 | 3 | 2 | 0 | 0 | || G=9| * public presentation Of The Kstar With keep To A examen physique For The White-wine Quality DatasetTesting mode| Training Set| Testing Set| 10-Fold Cross substantiation| 66% give out| Correctly sort out Instances| 99. 6581 %| 100 %| 70. 6624 %| 63. 9221 %| Kappa statistic| 0. 9949| 1| 0. 5365| 0. 4252| Mean Absolute wrongful conduct| 0. 0575| 0. 0788| 0. 1297| 0. 1379| Root Mean Squared error| 0. 1089| 0. one hundred forty-five| 0. 2428| 0. 2568| Relative Absolute Error| 29. 8022 %| | 67. 2423 %| 71. 2445 %| * The government issue Generated After Applying K-Star On Red-wine Quality Dataset Kstar Options : -B 70 -M a | clock Taken To Build Model: 0 Seconds| secern Cross- confirmation (10-Fold)| * Summary | Correctly classify Instances | 1013 | 71. 379 %| falsely classified ad Instances| 413 | 28. 9621 %| Kappa Statistic | 0. 5294| | Mean Absolute Error | 0. 1381| | Root Mean Squared Error | 0. 2592| | Relative Absolute Error | 64. 5286 %| | Root Relative Squared Error | 79. 309 %| | Total Number Of Instances | 1426 | | * Detailed true statement By Class | | TP commit | FP Rate | clearcutness | Recall | F-Measure | ROC Area | PRC Area| Class| | 0 | 0. 001 | 0 | 0 | 0 | 0. 574 | 0. 019 | 3| | 0 | 0. 003 | 0 | 0 | 0 | 0. 811 | 0. 114 | 4| | 0. 791| 0. 176 | 0. 67| 0. 791| 0. 779 | 0. 894 | 0. 867 | 5| | 0. 769 | 0. 26 | 0. 668 | 0. 769 | 0. 715 | 0. 834 | 0. 788 | 6| | 0. 511 | 0. 032 | 0. 692 | 0. 511 | 0. 588 | 0. 936 | 0. 722 | 7| | 0. 125 | 0. 001 | 0. 5 | 0. 125 | 0. 2 | 0. 896 | 0. 142 | 8| Weighted Avg. | 0. 71| 0. 184| 0. 685| 0. 71| 0. 693| 0. 871| 0. 78| | * Confusion Matrix | A | B | C | D | E | F| | Class| 0 | 1 | 4| 1 | 0 | 0 | | | A=3| 1 | 0 | 30| 17 | 0 | 0| | | B=4| 0 | 2| 477| one hundred twenty | 4 | 0| | | C=5| 0 | 1 | 103 | 444| 29 | 0| || D=6| 0 | 0 | 8 | 76 | 90 | 2 | || E=7| 0 | 0 | 0 | 7 | 7 | 2| || F=8| Performance Of The Kstar With measure To A Testing Configuration For The Red-wine Quality Dataset Testing order| Training Set| Testing Set| 10-Fold Cross Validation| 66% garbled| Correctly categorize Instances| 99. 7895 %| 100 % | 71. 0379 %| 70. 7216 %| Kappa statistic| 0. 9967| 1| 0. 5294| 0. 5154| Mean Absolute E rror| 0. 0338| 0. 0436| 0. 1381| 0. 1439| Root Mean Squared Error| 0. 0675| 0. 0828 | 0. 2592| 0. 2646| Relative Absolute Error| 15. 8067 %| | 64. 5286 %| 67. 4903 %| * J48 Decision Tree * Class for generating a pruned or unpruned C4. 5 decision tree. A decision tree is a prognosticative machine-learning model that decides the take aim value ( certified variable) of a new sample based on various attribute values of the available data. * The internal nodes of a decision tree denote the different attribute; the branches between the nodes tell us the possible values that these attributes can have in the observed samples, while the terminal nodes tell us the final value (classification) of the dependent variable. * The attribute that is to be predicted is known as the dependent variable, since its value depends upon, or is decided by, the values of all the other attributes.The other attributes, which help oneself in predicting the value of the dependent variable, are known as the in dependent variables in the dataset. * The J48 Decision tree classifier follows the following simple algorithm: * In order to clear a new item, it first needs to create a decision tree based on the attribute values of the available training data. So, whenever it encounters a set of items (training set) it identifies the attribute that discriminates the various instances most clearly. * This mark that is able to tell us most about the data instances so that we can classify them the best is said to have the highest information gain. at present, among the possible values of this feature, if there is any value for which there is no ambiguity, that is, for which the data instances falling within its mob have the same value for the stooge variable, then we terminate that branch and specialise to it the object lens value that we have obtained. * For the other cases, we then look for another attribute that gives us the highest information gain. Hence we quell in this manner until we either get a clear decision of what junto of attributes gives us a particular target value, or we run out of attributes.In the event that we run out of attributes, or if we cannot get an unambiguous result from the available information, we assign this branch a target value that the majority of the items under this branch possess. * Now that we have the decision tree, we follow the order of attribute selection as we have obtained for the tree. By checking all the respective attributes and their values with those countn in the decision tree model, we can assign or predict the target value of this new instance. * The Result Generated After Applying J48 On White-wine Quality Dataset era Taken To Build Model: 1. 4 Seconds| secern Cross-Validation (10-Fold) | * Summary| | | Correctly Classified Instances| 2740 | 58. 547 %| wrongly Classified Instances | 1940 | 41. 453 %| Kappa Statistic | 0. 3813| | Mean Absolute Error | 0. 1245| | Root Mean Squared Error | 0. 3194| | Relative Absol ute Error | 64. 5770 %| | Root Relative Squared Error| 102. 9013 %| | Total Number Of Instances | 4680| | * Detailed Accuracy By Class| | TP Rate| FP Rate| Precision| Recall| F-Measure| ROC Area| Class| | 0| 0. 002| 0| 0| 0| 0. 30| 3| | 0. 239| 0. 020| 0. 270| 0. 239| 0. 254| 0. 699| 4| | 0. 605| 0. 169| 0. 597| 0. 605| 0. 601| 0. 763| 5| | 0. 644| 0. 312| 0. 628| 0. 644| 0. 636| 0. 689| 6| | 0. 526| 0. 099| 0. 549| 0. 526| 0. 537| 0. 766| 7| | 0. 363| 0. 022| 0. 388| 0. 363| 0. 375| 0. 75| 8| | 0| 0| 0| 0| 0| 0. 496| 9| Weighted Avg. | 0. 585 | 0. 21 | 0. 582 | 0. 585 | 0. 584 | 0. 727| | * Confusion Matrix | A| B| C| D| E| F| G| || Class| 0| 2| 6| 5| 0| 0| 0| || A=3| 1| 34| 55| 44| 6| 2| 0| || B=4| 5| 50| 828| 418| 60| 7| 0| || C=5| 2| 32| 413| 1357| 261| 43| 0| || D=6| | 7| 76| 286| 459| 44| 0| || E=7| 1| 1| 10| 49| 48| 62| 0| || F=8| 0| 0| 0| 1| 2| 2| 0| || G=9| * Performance Of The J48 With Respect To A Testing Configuration For The White-wine Quality Dataset Testing manner| T raining Set| Testing Set| 10-Fold Cross Validation| 66% secernate| Correctly Classified Instances| 90. 1923 %| 70 %| 58. 547 %| 54. 8083 %| Kappa statistic| 0. 854| 0. 6296| 0. 3813| 0. 33| Mean Absolute Error| 0. 0426| 0. 0961| 0. 1245| 0. 1347| Root Mean Squared Error| 0. 1429| 0. 2756| 0. 3194| 0. 3397| Relative Absolute Error| 22. 0695 %| | 64. 577 %| 69. 84 %| * The Result Generated After Applying J48 On Red-wine Quality Dataset era Taken To Build Model: 0. 17 Seconds| Stratified Cross-Validation| * Summary| Correctly Classified Instances | 867 | 60. 7994 %| Incorrectly Classified Instances | 559 | 39. 2006 %| Kappa Statistic | 0. 3881| | Mean Absolute Error | 0. 1401| | Root Mean Squared Error | 0. 3354| | Relative Absolute Error | 65. 4857 %| | Root Relative Squared Error | 102. 602 %| |Total Number Of Instances | 1426 | | * Detailed Accuracy By Class| | Tp Rate | Fp Rate | Precision | Recall | F-measure | Roc Area | Class| | 0 | 0. 004 | 0 | 0 | 0 | 0. 573 | 3| | 0. 063 | 0. 037 | 0. 056 | 0. 063 | 0. 059 | 0. 578 | 4| | 0. 721 | 0. 258 | 0. 672 | 0. 721 | 0. 696 | 0. 749 | 5| | 0. 57 | 0. 238 | 0. 62 | 0. 57 | 0. 594 | 0. 674 | 6| | 0. 563 | 0. 64 | 0. 553 | 0. 563 | 0. 558 | 0. 8 | 7| | 0. 063 | 0. 006 | 0. 1 | 0. 063 | 0. 077 | 0. 691 | 8| Weighted Avg. | 0. 608 | 0. 214 | 0. 606 | 0. 608 | 0. 606 | 0. 718 | | * Confusion Matrix | A | B | C | D | E | F | | Class| 0 | 2 | 1 | 2 | 1 | 0 | | | A=3| 2 | 3 | 25 | 15 | 3 | 0 | | | B=4| 1 | 26 | 435 | 122 | 17 | 2 | | | C=5| 2 | 21 | 167 | 329 | 53 | 5 | | | D=6| 0 | 2 | 16 | 57 | 99 | 2 | | | E=7| 0 | 0 | 3 | 6 | 6 | 1 | | | F=8| Performance Of The J48 With Respect To A Testing Configuration For The Red-wine Quality Dataset Testing Method| Training Set| Testing Set| 10-Fold Cross Validation| 66% Split| Correctly Classified Instances| 91. 1641 %| 80 %| 60. 7994 %| 62. 4742 %| Kappa statistic| 0. 8616| 0. 6875| 0. 3881| 0. 3994| Mean Absolute Error| 0. 0461| 0. 0942| 0. 1401| 0. 1323| Root Mean Squared Er ror| 0. 1518| 0. 2618| 0. 3354| 0. 3262| Relative Absolute Error| 21. 5362 %| 39. 3598 %| 65. 4857 %| 62. 052 %| * Multilayer Perceptron * The back propagation algorithm performs learning on a multilayer feed-forward neural network. It iteratively learns a set of weights for prediction of the class label of tuples. * A multilayer feed-forward neural network consists of an stimulant signal layer, one or more unfathomable layers, and an produce layer. * Each layer is made up of building blocks. The inputs to the network correspond to the attributes measured for each training tuple. The inputs are fed at the same duration into the units making up the input layer. These inputs pass through the input layer and are then weighted and fed simultaneously to a second layer of â€Å"neuronlike” units, known as a hidden layer. The widenings of the hidden layer units can be input to another hidden layer, and so on. The number of hidden layers is arbitrary, although in practice, us ually only one is used. The weighted makes of the last hidden layer are input to units making up the output layer, which emits the network’s prediction for given tuples. * The units in the input layer are called input units. The units in the hidden layers and output layer are sometimes referred to as neurodes, due to their emblematic biological basis, or as output units. * The network is feed-forward in that none of the weights cycles back to an input unit or to an output unit of a previous layer.It is full connected in that each unit provides input to each unit in the next forward layer. * The Result Generated After Applying Multilayer Perceptron On White-wine Quality Dataset Time taken to build model: 36. 22 seconds| Stratified cross-validation| * Summary| Correctly Classified Instances | 2598 | 55. 5128 %| Incorrectly Classified Instances | 2082 | 44. 4872 %| Kappa statistic | 0. 2946| | Mean absolute error | 0. 1581| | Root mean squared error | 0. 2887| |Relative absolu te error | 81. 9951 %| | Root relative squared error | 93. 0018 %| | Total Number of Instances | 4680 | | * Detailed Accuracy By Class | | TP Rate | FP Rate | Precision | Recall | F-Measure | ROC Area | PRC Area | Class| | 0 | 0 | 0 | 0 | 0 | 0. 344 | 0. 002 | 3| | 0. 056 | 0. 004 | 0. 308 | 0. 056 | 0. 095 | 0. 732 | 0. 156 | 4| | 0. 594 | 0. 165 | 0. 597 | 0. 594 | 0. 595 | 0. 98 | 0. 584 | 5| | 0. 704 | 0. 482 | 0. 545 | 0. 704 | 0. 614 | 0. 647 | 0. 568 | 6| | 0. 326 | 0. 07 | 0. 517 | 0. 326 | 0. 4 | 0. 808 | 0. 474 | 7| | 0. 058 | 0. 002 | 0. 5 | 0. 058 | 0. one hundred five | 0. 8 | 0. 169 | 8| | 0 | 0 | 0| 0 | 0 | 0. 356 | 0. 001 | 9| Weighted Avg. | 0. 555 | 0. 279 | 0. 544 | 0. 555 | 0. 532 | 0. 728 | 0. 526| | * Confusion Matrix |A | B | C | D | E | F | G | | Class| 0 | 0 | 5 | 7 | 1 | 0 | 0 | | | A=3| 0 | 8 | 82 | 50 | 2 | 0 | 0 | | | B=4| 0 | 11 | 812 | 532 | 12 | 1 | 0 | | | C=5| 0 | 6 | 425 | 1483 | 188 | 6 | 0 | | | D=6| 0 | 1 | 33 | 551 | 285 | 3 | 0 | | | E=7| 0 | 0 | 3 | 98 | 60 | 10 | 0 | | | F=8| 0 | 0 | 0 | 2 | 3 | 0 | 0 | | | G=9| * Performance Of The Multilayer perceptron With Respect To A Testing Configuration For The White-wine Quality DatasetTesting Method| Training Set| Testing Set| 10-Fold Cross Validation| 66% Split| Correctly Classified Instances| 58. 1838 %| 50 %| 55. 5128 %| 51. 3514 %| Kappa statistic| 0. 3701| 0. 3671| 0. 2946| 0. 2454| Mean Absolute Error| 0. 1529| 0. 1746| 0. 1581| 0. 1628| Root Mean Squared Error| 0. 2808| 0. 3256| 0. 2887| 02972| Relative Absolute Error| 79. 2713 %| | 81. 9951 %| 84. 1402 %| * The Result Generated After Applying Multilayer Perceptron On Red-wine Quality Dataset Time taken to build model: 9. 14 seconds| Stratified cross-validation (10-Fold)| * Summary | Correctly Classified Instances | 880 | 61. 111 %| Incorrectly Classified Instances | 546 | 38. 2889 %| Kappa statistic | 0. 3784| | Mean absolute error | 0. 1576| | Root mean squared error | 0. 3023| | Relative absolute error | 73. 6593 %| | Root relative squared error | 92. 4895 %| | Total Number of Instances | 1426| | * Detailed Accuracy By Class | | TP Rate | FP Rate | Precision | Recall | F-Measure | ROC Area | Class| | 0 | 0 | 0 | 0 | 0 | 0. 47 | 3| | 0. 42 | 0. 005 | 0. 222 | 0. 042 | 0. 070 | 0. 735 | 4| | 0. 723 | 0. 249 | 0. 680 | 0. 723 | 0. 701 | 0. 801 | 5| | 0. 640 | 0. 322 | 0. 575 | 0. 640 | 0. 605 | 0. 692 | 6| | 0. 415 | 0. 049 | 0. 545 | 0. 415 | 0. 471 | 0. 831 | 7| | 0 | 0 | 0 | 0 | 0 | 0. 853 | 8| Weighted Avg. | 0. 617 | 0. 242 | 0. 595 | 0. 617 | 0. 602 | 0. 758| | * Confusion Matrix | A | B | C | D | E | F | | Class| | 0 | 5 | 1 | 0 | 0| || A=3| 0 | 2 | 34 | 11 | 1 | 0 | | | B=4| 0 | 2 | 436 | 160 | 5 | 0 | | | C=5| 0 | 5 | 156 | 369 | 47 | 0 | | | D=6| 0 | 0 | 10 | 93 | 73 | 0 | | | E=7| 0 | 0 | 0 | 8 | 8 | 0 | | | F=8| * Performance Of The Multilayer perceptron With Respect To A Testing Configuration For The Red-wine Quality Dataset Testing Method| Training Set| Testing Set| 10-Fold Cross Va lidation| 66% Split| Correctly Classified Instances| 68. 7237 %| 70 %| 61. 7111 %| 58. 7629 %| Kappa statistic| 0. 4895| 0. 5588| 0. 3784| 0. 327| Mean Absolute Error| 0. 426| 0. 1232| 0. 1576| 0. 1647| Root Mean Squared Error| 0. 2715| 0. 2424| 0. 3023| 0. 3029| Relative Absolute Error| 66. 6774 %| 51. 4904 %| 73. 6593 %| 77. 2484 %| * Result * The classification experiment is measured by accuracy percentage of classifying the instances correctly into its class according to quality attributes ranges between 0 (very bad) and 10 (ex carrellent). * From the experiments, we institute that classification for red wine quality using Kstar algorithm achieved 71. 0379 % accuracy while J48 classifier achieved about 60. 7994% and Multilayer Perceptron classifier achieved 61. 7111% accuracy. For the white wine, Kstar algorithm yielded 70. 6624 % accuracy while J48 classifier yielded 58. 547% accuracy and Multilayer Perceptron classifier achieved 55. 5128 % accuracy. * Results from the expe riments lead us to conclude that Kstar performs better in classification task as compared against the J48 and Multilayer Perceptron classifier. The processing time for Kstar algorithm is also observed to be more efficient and less time consuming despite the large coat of wine properties dataset. 7. COMPARISON OF DIFFERENT ALGORITHM * The Comparison Of All ternion Algorithm On White-wine Quality Dataset (Using 10-Fold Cross Validation) Kstar| J48| Multilayer Perceptron| Time (Sec)| 0| 1. 08| 35. 14| Kappa Statistics| 0. 5365| 0. 3813| 0. 29| Correctly Classified Instances (%)| 70. 6624| 58. 547| 55. 128| True positive Rate (Avg)| 0. 707| 0. 585| 0. 555| False affirmative Rate (Avg)| 0. 2| 0. 21| 0. 279| * Chart Shows The Best suited Algorithm For Our Dataset (Measures Vs Algorithms) * In above chart, equation of True Positive rate and kappa statistics is given against three algorithm Kstar, J48, Multilayer Perceptron * Chart describes algorithm which is best suits for our datase t. In above chart column of TP rate & Kappa statistics of Kstar algorithm is high than other two algorithms. * In above chart you can see that the False Positive Rate and the Mean Absolute Error of the Multilayer Perceptron algorithm is high compare to other two algorithms. So it is not good for our dataset. * But for the Kstar algorithm these two values are less, so the algorithm having lowest values for FP Rate & Mean Absolute Error rate is best suited algorithm. * So the final we can make proof that the Kstar algorithm is best suited algorithm for White-wine Quality dataset. The Comparison Of All Three Algorithm On Red-wine Quality Dataset (Using 10-Fold Cross Validation) | Kstar| J48| Multilayer Perceptron| Time (Sec)| 0| 0. 24| 9. 3| Kappa Statistics| 0. 5294| 0. 3881| 0. 3784| Correctly Classified Instances (%)| 71. 0379| 60. 6994| 61. 7111| True Positive Rate (Avg)| 0. 71| 0. 608| 0. 617| False Positive Rate (Avg)| 0. 184| 0. 214| 0. 242| * For Red-wine Quality data set have also Kstar is best suited algorithm , because of TP rate & Kappa statistics of Kstar algorithm is higher than other two algorithms and FP rate & Mean Absolute Error of Kstar algorithm is lower than other algorithms. . APPLYING TESTING DATASET graduation1: Load pre-processed dataset. footstep2: Go to classify tab. Click on choose spill and select lazy leaflet from the hierarchy tab and then select kstar algorithm. After selecting the kstar algorithm keep the value of cross validation = 10, then build the model by clicking on swallow button. Step3: Now take any 10 or 15 records from your dataset, make their class value unknown(by putting ’? ’ in the cell of the corresponding raw ) as shown below. Step 4: Save this data set as . rff file. Step 5: From â€Å"test option” control board select â€Å"supplied test set”, click on to the set button and open the test dataset file which was lastly created by you from the disk. Step 6: From â⠂¬Å"Result list panel” panel select Kstar-algorithm (because it is better than any other for this dataset), right click it and click â€Å"Re-evaluate model on current test set” Step 7: Again right click on Kstar algorithm and select â€Å"visualize classifier error” Step 8:Click on save button and then save your test model.Step 9: After you had saved your test model, a separate file is created in which you go forth be having your predicted values for your testing dataset. Step 10: Now, this test model testament have all the class value generated by model by re-evaluating model on the test data for all the instances that were set to unknown, as shown in the reckon below. 9. ACHIEVEMENT * Classification models may be used as part of decision support system in different stages of wine production, hence giving the luck for manufacturer to make corrective and additive measure that will result in higher quality wine macrocosm produced. From the resulting classific ation accuracy, we found that accuracy rate for the white wine is influenced by a higher number of physicochemistry attribute, which are alcohol, density, plain sulfur dioxide, chlorides, citric acid, and volatile acidity. * Red wine quality is highly match to only four attributes, which are alcohol, sulphates, total sulfur dioxide, and volatile acidity. * This shows white wine quality is affected by physicochemistry attributes that does not affect the red wine in general. Therefore, I suggest that white wine manufacturer should conduct wider range of test particularly towards density and chloride content since white wine quality is affected by such substances. * Attribute selection algorithm we conducted also ranked alcohol as the highest in both datasets, hence the alcohol level is the main attribute that determines the quality in both red and white wine. * My suggestion is that wine manufacturer to focus in maintaining a suitable alcohol content, may be by eternal fermentation period or higher yield fermenting yeast.\r\n'