Improving JP with Improvised Prediction Model

Yesterday I wrote on low ridership (read here), of which out of many factors, information availability for journey planning contributes 12% from overall factors. However, the value is based on a survey on one location; that is Penang. Nonetheless, I believed, information availability is the key importance for a smooth journey planning, no matter of the services used or the impact to ridership. The reason for this is based on comments in Google Play for various Journey Planner such as Moovit, Transit Apps, SWIVL, SITS, etc, whereby many user stated their frustration on the accuracy of information displayed by the apps.

Beside information availability being a vital role for riders to plan their journey and services to use, based on the articles referred to in the post, the information must be also reliable, accurate and at real-time (or at least near real-time). If you received an information that was accurate a few minutes ago, there is a probability of the information to be inaccurate at the time of view or receive resulting in the inaccurate plan and action.

How to improve information accuracy, no matter how and when the information arrives at the user?
I did a quick review too on the following articles:

1. https://www.papercast.com/insights/predict-accurate-bus-arrival-journey-times/
2. https://core.ac.uk/download/pdf/82293981.pdf
3. https://escholarship.org/uc/item/51t364vz
4. http://gamma.cs.unc.edu/TROUTE/
5. https://www.researchgate.net/publication/274028208_Multimodal_Public_Transit_Trip_Planner_with_Real-Time_Transit_Data
6. https://repositorio-aberto.up.pt/bitstream/10216/6817/2/26915.pdf
7. https://jungleworks.com/predicting-accurate-arrival-time/
8. https://www.researchgate.net/publication/332342499_Survey_of_ETA_prediction_methods_in_public_transport_networks
9. https://dl.acm.org/citation.cfm?id=3219819.3219874
10. https://pdfs.semanticscholar.org/8c95/f20cd049e5f0d35466544958631e3e10c258.pdf
11. https://www.papercast.com/wp-content/uploads/2017/06/Papercast_A4_Better-ETA_2017.pdf
12. https://datascience.stackexchange.com/questions/10301/how-to-predict-eta-using-regression
13. https://ieeexplore.ieee.org/abstract/document/1212964
14. https://www.tandfonline.com/doi/abs/10.1080/15472450600981009
15. https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1467-8667.2004.00363.x
16. https://journals.sagepub.com/doi/abs/10.3141/1666-12
17. https://link.springer.com/chapter/10.1007/978-981-13-3393-4_29
18. https://patents.google.com/patent/US10254119B2/en
19. https://arxiv.org/abs/1904.05037
20. https://patents.google.com/patent/US20190130260A1/en
21. https://www.tandfonline.com/doi/abs/10.1080/19427867.2017.1366120
22. https://link.springer.com/article/10.1007/s12652-019-01198-1
23. https://arxiv.org/abs/1904.03444
24. https://patents.google.com/patent/US20190051154A1/en
25. https://ieeexplore.ieee.org/abstract/document/8691701

Based on the articles above, below is the gist of the findings:

1. Reliable and accurate information at real-time (or near real-time) is critical for smooth journey planning
2. Recent technology (BDA, AI, ML & IoT) has resulted in many new algorithms to predict accurate ETA to be used in JP
3. The most recent is KNN which requires lots of historical data and real-time tracking for accurate prediction

Recommendation:

1. To research and try-n-error all the algorithms to find the best algorithm to predict accurate ETA to be used in the Malaysian environment
2. To define the best algorithm based on time of request (peak or non-peak)
3. To come out with a new algorithm and flow to ensure the ETA for JP is 95% accuracy during non-peak and 90% accuracy during peak.