BUS ROUTE DEMAND PREDICTION WITH DEEP LEARNING

Lai, Stevanus Alditian (2022) BUS ROUTE DEMAND PREDICTION WITH DEEP LEARNING. Other thesis, Universitas Katholik Soegijapranata Semarang.

Preview	Text 18.K1.0084-Stevanus Alditian Lai_COVER_a.pdf Download (2MB) | Preview
Preview	Text 18.K1.0084-Stevanus Alditian Lai_BAB I_a.pdf Download (2MB) | Preview
	Text 18.K1.0084-Stevanus Alditian Lai_BAB II_a.pdf Restricted to Registered users only Download (2MB)
Preview	Text 18.K1.0084-Stevanus Alditian Lai_BAB III_a.pdf Download (2MB) | Preview
Preview	Text 18.K1.0084-Stevanus Alditian Lai_BAB IV_a.pdf Download (2MB) | Preview
Preview	Text 18.K1.0084-Stevanus Alditian Lai_BAB V_a.pdf Download (2MB) | Preview
Preview	Text 18.K1.0084-Stevanus Alditian Lai_BAB VI_a.pdf Download (2MB) | Preview
Preview	Text 18.K1.0084-Stevanus Alditian Lai_DAPUS_a.pdf Download (2MB) | Preview
Preview	Text 18.K1.0084-Stevanus Alditian Lai_LAMP_a.pdf Download (2MB) | Preview

Abstract

bus companies currently have several obstacles in providing their fleets from one city to another because of the highly dynamic demand from passengers, bus companies must be able to analyze which routes will have a lot of demand so that bus companies can provide more fleets on the routes that will have high demand. Unfortunately the bus company is currently still unable to predict which routes will be in high demand, at this time the bus company can only guess. Currently, to overcome this, the bus company has collected data which will later be analyzed. Since the deep learing method is relatively new for bus company to predict the bus route demand, this study explores new method to make the bus company more profitable by trying to create and implement LSTM Autoencoder-Bi-LSTM Hybrid Models and Bi-LSTM to forcast bus route demand to support the decision making process in orrder to optimize bus fleet deployment each route. The results shows that LSTM Autoencoder-Bi-LSTM Hybrid Models and Bi-LSTM models doesn't differ very much, the loss and metrcs value differ a little, and both models performs quiet well, but 1 things that differ these 2 models, that is the training time, the autoencoders training time is very slow compared to models without autoencoders. This is normal for autoencoder to train slower than without it due to more network depth of the models with autoencoder.

Item Type:	Thesis (Other)
Subjects:	000 Computer Science, Information and General Works
Divisions:	Faculty of Computer Science > Department of Informatics Engineering
Depositing User:	mr AM. Pudja Adjie Sudoso
Date Deposited:	23 Mar 2022 04:12
Last Modified:	23 Mar 2022 04:12
URI:	http://repository.unika.ac.id/id/eprint/28280

Actions (login required)

View Item