Ghazali, Nurul Bashirah (2022) Cable fault detection for DSL communication using machine learning. Masters thesis, Universiti Tun Hussein Onn Malaysia.

Preview	Text 24p NURUL BASHIRAH GHAZALI.pdf Download (644kB) | Preview
	Text (Copyright Declaration) NURUL BASHIRAH GHAZALI COPYRIGHT DECLARATION.pdf Restricted to Repository staff only Download (378kB) | Request a copy
	Text (Full Text) NURUL BASHIRAH GHAZALI WATERMARK.pdf Restricted to Registered users only Download (34MB) | Request a copy

Abstract

Very-High-Bit-Rate Digital Subscriber Line 2 (VDSL2) is a technology that is widely deployed all over the world, but its performance may be limited due to its nature, thus causing it to be more susceptible to signal degradation and faulty lines. Some of the VDSL2 line conditions are ideal, bridge tap, partial-short, short-wired, partial-open and open-wired faults. However, VDSL2 is still one of the technologies that users in sub-urban areas depend on to access the internet service. This thesis aimed to demonstrate the classification of VDSL2 line impairments and the regression of VDSL2 line-impairment locations using a machine learning algorithm. VDSL2 line-impairment classification and localisation detection may increase the technology’s overall performance. The proposed system was developed using Python programming language following the stages for classification and regression using the Random Forest machine learning algorithm. The proposed algorithm was selected based on comparisons of several algorithms’ performances in the WEKA software. The first stage involved data acquisition from the Multi-Service Access Node (MSAN) in the VDSL2 laboratory setup. The laboratory was set up to imitate the actual VDSL2 network. Due to certain VDSL2 line impairments that could cause the unavailability of some attributes, the acquired data needed to undergo the pre-processing stage. The training of the pre-processed data was evaluated using the percentage-split method. The classification of VDSL2 line impairments was assessed based on the percentage of accuracy and correctly classified instances. Using the proposed method, the developed system for the classification and localisation of VDSL2 line impairments was able to achieve 97.73% accuracy and a correlation coefficient of 0.9933 for the VDSL2 classification and localisation models, respectively, after using same dataset as that for training. Validation using the latest daily VDSL2 dataset was able to achieve 100% accuracy and a correlation coefficient of 0.36

Actions (login required)

View Item