Please use this identifier to cite or link to this item:
|Title:||Machine Learning Techniques In Categorical Time Series Analysis Of Manufacturing Process|
Ισιδώρα Χαρά Τουρνή
variable length classification
categorical time series
|Abstract:||The complexity of modern industrial processes and the constant innovations in production monitoringtechnologies and data collection, strongly outline the need for advancements in productiondata analysis. Data Mining is a rapidly growing field, aiming in understanding data and extractingpreviously unknown information, with the use of Machine Learning techniques, in order to optimizeproduction.Our cooperation with Johnson & Johnson, enabled us to obtain and explore real case productiondata. The exploitation of them focused on two distinct goals. The first one was the visualization of thedata and the graphical representation of all variables characterizing the mixing process, for an improveddata overview. The second one was the Machine Learning algorithms’ modification and applicationon the properly pre-processed production data, to look into the possibilities of these techniques in theenterprise space.Machine Learning, by definition, sets to represent data as objects in space, utilizing labels anddistances between them. In this direction, data were grouped into objects and vectorized with varioustechniques. Classification and Clustering algorithms were parameterized and implemented, investigatingunique attributes of the provided data. Distance calculation methods for each algorithm wereexamined in depth, and each experiment was assessed through different evaluation metrics, in order toexamine the performance of the algorithms and the result’s accuracy, compared to the initial data. Ourconclusions indicate that Machine Learning can drive important business decisions, through processquantification, and further research can be done in each specific case.|
|Appears in Collections:||Διπλωματικές Εργασίες - Theses|
Items in Artemis are protected by copyright, with all rights reserved, unless otherwise indicated.