Defining a Machine Learning implementation for demand forecasting in Dental Tracking System
Macias Vargas, Camilo (2019)
Macias Vargas, Camilo
Åbo Akademi
2019
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe201902236082
https://urn.fi/URN:NBN:fi-fe201902236082
Tiivistelmä
In this document we conduct a comparison between the non-linear SVR (Support Vector Regressor) and the LSTM (Long-Short-Term Memory) Recurrent Neural models, for the prediction of dental assets demand on institutions using the DTS (Dental Tracking System) software developed by the company LM-Instruments. DTS is a Java programming language application that tracks usage of dental assets based on RFID (Radio Frequency Identification) technology. This work describes the generation of the data in DTS, defines what data is considered relevant to be used in the demand forecast; and presents the process of data preparation, as well as the development of the models and the validation process. Finally, the best results are presented, analysed and compared. This comparison reviews the performance, implementation simplicity and computational power required. After the work performed, the suggestion for LMInstruments is to use SVR with a polynomial kernel of degree one, using low penalty, epsilon and coefficient values. This implementation proved to have the best performance when using weekly based data sets over daily data sets.