Loading...
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 51048 (01)
- University: Sharif University of Technology
- Department: Industrial Engineering
- Advisor(s): Khedmati, Majid; Akhavan Niaki, Taghi
- Abstract:
- One of the most trivial factors for every service company is to fulfill customer demands and to satisfy them. For this purpose, it is needed to explore the business fundamentals. Transportation companies, however seem to be more important, because in one hand, the demand for their services is too many. In the other hand due to the intense competition between rivals, every business tries to stand in the crowd; thus, it is of high importance to correctly predict the travel time as well as fare. The importance of time is not only because of having a fare pricing system, but also it is important for scheduling, assigning drivers, and covering city or a particular district. Travel time information can help a business to have quality services and schedule their workforce through the day.In this research, town trips are studied. Data mining and statistical techniques are used in order to extract the knowledge of interest. Main purposes for this study are identiying important factors, grouping categorical levels, clustering travels and at the end, predicting travel time duration. In this study, not only significant features are identified with ANOVA, but also two novel approaches, mathematical and statistical, are proposed which are able to group the levels of a categorical variable. Then at the end, the effect of each is traced in the predictive performance of relevant models.In the prediction section, artificial neural network models are trained separately for both short and long trips which will be identified in advance. For this purpose, a random forest model is trained to predict whether a new observation will be categorized as a short or long trip. Then, based on the result, relevant trained neural network model will be used to predict the duration time for an unseen observation. In this part, the effect of mathematical and statistical approaches are also explored ,and as the result, the mathematical modeling is selected due to its better performance. The test RMSE for ANN model for short trips with mathematical modeling for weather and hour variables, was 4.23 minutes.At last, a new predictive model is proposed, modified nearest neighbourhoods, which finds the most similar observations from past in multiple stages. Then according to the average duration for those selected observations, it gives estimation for time duration. Test RMSE for this model was 4.4 minutes which shows higher predictive power in comparison with previous studies
- Keywords:
- Prediction ; Clustering ; Artificial Neural Network ; Travel Time ; Customer Satisfaction ; Town Trip
Digital Object List
-
محتواي کتاب
- view
Bookmark