Classification of Quality Problems in Carpet Manufacturing by Using Data Mining Algorithms

UNUTMAZ DURMUŞOĞLU Z. D., Borsöken A.

1st Central American and Caribbean International Conference on Industrial Engineering and Operations Management, IEOM 2021, Port-Au-Prince, Haiti, 15 - 16 Haziran 2021, ss.25-34, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Basıldığı Şehir: Port-Au-Prince
  • Basıldığı Ülke: Haiti
  • Sayfa Sayıları: ss.25-34
  • Anahtar Kelimeler: Classification Methods, Data mining (DM), Machine Learning (ML), Statistical Analysis, Waikato Environment for Knowledge Analysis (WEKA) Software
  • Samsun Üniversitesi Adresli: Hayır

Özet

Nowadays, with the latest technology, the accumulation of data has increased, and it has become more important to transform the stored raw data into information. The transformation of raw data into information increases earnings and can more easily be adapted to the competitive environment. Data mining is an approach based on statistical applications and machine learning algorithms in converting raw data into information. Data mining has been an important research area to find the hidden information/knowledge inside huge amounts of data. In this paper, quality problems in carpet manufacturing are gathered in a database and the data are analyzed and classified by using data mining algorithms. The main purpose is to classify the quality problems in carpet production. The similarity of the problems resulted in the grouping of them under the same title. Thus, a deeper perspective on carpet quality problems is targeted and the number of customer complaints is expected to reduce in the long term. The data obtained from the database of a carpet producer were pre-processed and adapted to WEKA 3.9.4 software. The dataset was classified using 10-fold cross-validation and Percentage-Split with J.48, two Bayesian classifiers (NaiveBayes and BayesNet), a Nearest Neighbour algorithm (IBk), and two rule learners (OneR and JRip). The achieved conclusions present that the decision tree classifier (J48) performs best (with the highest overall accuracy), followed by the rule learner (JRip) and the BayesNet classifier. The k-NN classifiers are less accurate than the others. In consequence of the J48 classifier, the True Positive Rate is high for three of the classes – Bad, Good, and Average, while it is very low for the other two classes – Very Good and Excellent. The Precision is very high for the Bad class, high for the Good and Average classes, and low for the Very Good and Excellent classes. The acquired results are a little better for the 10-fold Cross-Validation testing option.