Advanced

Investigation of Machine Learning Applicability on Mobile Device Diagnostics for Quality Prediction

Dagasan, Edin LU and Josefsson, Rasmus LU (2016) In Master's Theses in Mathematical Sciences FMA820 20161
Mathematics (Faculty of Engineering)
Abstract
Sony Mobile Communications (SoMC) perpetually collect large amounts of prototype device data into their database which is used for analysing their devices. Different areas are analysed individually using this data but no general model exists which encapsulates the overall performance. A desired model would include information from the different key performance areas and utilise machine learning with the objective to prognosticate the prototype device quality.

This thesis investigates the possibility of using machine learning tools to generalise the data analysis and build a predictive quality model based on historical data. It consists of extracting and processing data from the company database, finding a relevant feature representation... (More)
Sony Mobile Communications (SoMC) perpetually collect large amounts of prototype device data into their database which is used for analysing their devices. Different areas are analysed individually using this data but no general model exists which encapsulates the overall performance. A desired model would include information from the different key performance areas and utilise machine learning with the objective to prognosticate the prototype device quality.

This thesis investigates the possibility of using machine learning tools to generalise the data analysis and build a predictive quality model based on historical data. It consists of extracting and processing data from the company database, finding a relevant feature representation and evaluating different models based on both supervised and unsupervised learning methods. The study builds on the assumption that performance is related to the device software and its development, through which conclusions about quality could then be made.
The results show that finding relevant features that distinguish different device software is very difficult, rendering an unsupervised approach inadequate. Apart from deficient features, the supervised methods suffer from the unreliability of using survey data as labels. In conclusion, no explicit model is recommended and different approaches are preferable where another feature representation and more qualitative labels are the main requirements. (Less)
Please use this url to cite or link to this publication:
author
Dagasan, Edin LU and Josefsson, Rasmus LU
supervisor
organization
course
FMA820 20161
year
type
H2 - Master's Degree (Two Years)
subject
keywords
big data, machine learning, data mining, mobile device diagnostics
publication/series
Master's Theses in Mathematical Sciences
report number
LUTFMA-3295-2016
ISSN
1404-6342
other publication id
2016:E22
language
English
id
8881112
date added to LUP
2016-08-25 13:40:37
date last changed
2016-08-25 15:49:45
@misc{8881112,
  abstract     = {Sony Mobile Communications (SoMC) perpetually collect large amounts of prototype device data into their database which is used for analysing their devices. Different areas are analysed individually using this data but no general model exists which encapsulates the overall performance. A desired model would include information from the different key performance areas and utilise machine learning with the objective to prognosticate the prototype device quality.

This thesis investigates the possibility of using machine learning tools to generalise the data analysis and build a predictive quality model based on historical data. It consists of extracting and processing data from the company database, finding a relevant feature representation and evaluating different models based on both supervised and unsupervised learning methods. The study builds on the assumption that performance is related to the device software and its development, through which conclusions about quality could then be made.
The results show that finding relevant features that distinguish different device software is very difficult, rendering an unsupervised approach inadequate. Apart from deficient features, the supervised methods suffer from the unreliability of using survey data as labels. In conclusion, no explicit model is recommended and different approaches are preferable where another feature representation and more qualitative labels are the main requirements.},
  author       = {Dagasan, Edin and Josefsson, Rasmus},
  issn         = {1404-6342},
  keyword      = {big data,machine learning,data mining,mobile device diagnostics},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master's Theses in Mathematical Sciences},
  title        = {Investigation of Machine Learning Applicability on Mobile Device Diagnostics for Quality Prediction},
  year         = {2016},
}