000			05063ntm a2200361 i 4500
999			_c95408 _d95414
003			MY-KuUP
005			20251125105845.0
006			t||||fr|||| 000 0
008			220303s2020 my a|||fr|||| 000 0 eng d
020			_aTHE0009133(Local) _qhardback
040			_aUMP _beng _cUMP _erda
090			_aFTKPM .C46 2020 r Thesis
100	1		_aChoong Chun Sern, _eauthor.
245	1	0	_aAutomatic detection and indication of pallet-level tagging from rfid readings using machine learning algorithms / _cChoong Chun Sern
264		1	_aKuantan, Pahang : _bUMP, _c2020
264		4	_c© 2020
300			_axvii, 171 pages : _billustrations (some color) ; _c30 cm. + _e1 CD ROM
336			_atext _2rdacontent
336			_atext _2rdacontent
337			_aunmediated _2rdamedia
337			_acomputer _2rdamedia
338			_avolume _2rdacarrier
338			_acomputer disc _2rdacarrier
347			_atext file _bPDF _2rda
500			_aFaculty of Manufacturing and Mechatronic Engineering Technology
502			_aThesis (Master of Science) -- Universiti Malaysia Pahang – 2020
504			_aIncludes bibliographical references
520	3		_aIdentifying specific locations of items such as containers, warehouse pellets, and returnable packages in a large environment, for instance, in a warehouse, requires an extensive tracking system that could identify the location through data visualization. This is the similar case for radio-frequency identification (RFID) pallet level signal as the accuracy of determining the position for specific location either on the level or stacked in the same direction are read uniformly. However, there is no single study focusing on pallet-level classification, in particular on distance measurement of pallet height. Hence, a methodological approach that could provide the solution is essential to reduce the misplaced issues and thus reduce the problem in searching the products in a large-scale setting. The objective of this work attempts to define the pallet level of the stacked RFID tags through the machine learning techniques framework. The methodology started with the pallet-level which firstly determined by manual clustering according to the product code number of the tags that were manufactured for defining the actual level. An additional study of the radio frequency of the tagged pallet box in static condition was carried out by determining the feature of the time series. Various sample sizes of 1 Hz, 5 Hz and 10 Hz combined with the received signal strength of maximum, minimum, mode, median, mean, variance, maximum and minimum difference, kurtosis and skewness are evaluated. The statistical features of the received signal strength reading are analyzed by the selection of the univariate features, feature importance technique, and principal component analysis. The received signal strength of the maximum, median, and mean of all statistical features has been shown to be significant specifically for the 10Hz sample size. Different machine learning classifiers were tested based on the significant features, namely the Artificial Neural Network, Decision Tree, Random Forest, Naive Bayes Support Vector Machine, and k-Nearest Neighbors. It was shown that up to 95.02% of the trained Random Forest Model could be classified, indicating that the established framework is viable for pallet classification. Furthermore, the efficacy of different models based on heuristic hyperparameter tuning is evaluated in which the different kernel function for Support Vector Machine, various distance metrics of k-Nearest Neighbors. The ensemble learning technique, changes of activation function in Neural Network as well as the unsupervised learning (k-means clustering algorithm and Friis Transmission Equation) was also applied to classify the multiclass classification in pallet-level. In results, it was found that the Random Forest provided 92.44% of the test sets with the highest accuracy. In order to further validate the position of the tagging in the pallet box of the Random Forest model developed, a different predefined location was used to validate the model. The best position that could achieve a classification accuracy of 93.30% through the validation process for position five (5) in the systematic model that is the centre of the pallet box. In conclusion, it can be inferred from the analysis that the Random Forest model has better predictive performance compared to the rest of the pallet level partition model with a height of 12 cm used in this research. Based on the train, validation, and test sets in Random Forest, the RFID capability to determine the position of the pallet can be detected precisely.
610	2	0	_aFaculty of Manufacturing and Mechatronic Engineering Technology _xDissertations
650		0	_aUniversities and colleges _xDissertations
650		0	_aTheses
942			_2lcc _cTHESIS