Electric bus arrival and charging station placement assessment using machine learning techniques

Figures & data

Table 1. Summary of literatures pertaining to bus arrival prediction.

Studies	ML algorithms used	Best performing algorithm	Performance
Beshah and Hill (2010)	DT,NB,KNN	KNN	AUC-0.965 Accuracy-80.8%
Gu et al. (2021)	AOSVM, KNN, LR and ANN	AOSVM	MAPE- 9.9%
Liu, Sun, and Wang (2020)	RNN-LSTM, KR, KNN Delay and Additive	RNN-LSTM	LSTM > KR > k-NN > Delay > Additive
Wang (2021)	SVR, KNN	SVR+KNN	Maximum error of 84 s, a minimum error of 10 s, an average error of 42.5 s, average relative error of 5.74%.
Shanthi et al. (2022)	SVR, RFR, DTR, KNNR,GBR	SVR	Error value of 30 secs
Zhou et al. (2019)	RNN, SVM, MLP	DARNN	MAE and RMSE of DARNN is better than LSTM-RNN
Prakash et al. (2019)	DT,GEP-TS, GEP-REG,ANN-TS,ANN-REG and MLR	Time series ANN	MAE-10, RMSE-19
Li et al. (2017)	Three stage mixed model involving KNN, Kalman Filter and Markov model	Single step model	MAPE − 0.3
Treethidtaphat et al. (2017)	Deep neural network(DNN)	DNN	MAPE-0.55
Ou (2022)	LSTM	LSTM	MAE- 0.5 min, RMSE-0.72 min
Fan and Gurmu (2015)	Historical average, Kalman Filtering and ANN	ANN	Average MAPE- 8.5%

Figure 1. Data collection scheme in the study area.

A snapshot of data collection framework for the study area.

Figure 2. Bus route in the city.

Complete Bus Route in Addis Ababa city.

Table 2. Description of the selected features.

No.	Features Name	Data Type	Description
1	Beginning Date	Numerical	Initial Date of the trip
2	Beginning Time	Numerical	Initial Time of the trip
3	End Date	Numerical	End Date of the trip
4	Peak/off peak	Categorical	Shows the time of crowding/traffic congestion
5	Mileage	Numerical	Total distance in kilometer of the trip
6	Duration	Numerical	The difference between the projected arrival time and the actual arrival time
7	Initial latitude	Numerical	Source Latitude of vehicle
8	Initial longitude	Numerical	Source Latitude of vehicle
9	Final latitude	Numerical	Destination Latitude of vehicle
10	Final longitude	Numerical	Destination Latitude of vehicle
11	End Time	Numerical	End Time of the trip

Figure 3. Feature correlation (heat map).

Features present in the data and correlation between them.

Figure 4. Proposed machine learning model.

Proposed framework for testing with individual and ensemble-based regression models.

Table 3. Description of charging point selection steps.

Sl. No of Steps	Descriptions
Step 1. Data Collection	Gather dataset with route information
Step 2. Data Preprocessing	Clean, format date/time, ensure coordinates are numeric
Step 3. Exploratory Data Analysis (EDA)	Gain insights from dataset
Step 4 Feature Engineering	Create useful features if needed.
Step 5 Clustering Algorithm Selection	Choose K-Means, DBSCAN, or SOM.
Step 6 Apply Clustering Algorithms	Identify potential charging points.
Step 7 Visualize Clusters	Plot routes and clusters
Step 8 Select Optimal Points	Identify strategic locations based on clustering.
Step 9 Additional Filtering	Apply extra criteria.
Step 10 Map Visualization	Plot charging points on real map.
Step 11 Evaluation and Validation	Assess charging points based on criteria.
Step 12 Final Selection	Choose optimal EV charging points
Step 13 Generate Reports	Summarize process, criteria, and recommendations.

Figure 5. Dataset after adding average speed.

A snapshot of the Dataset with average speed attribute.

Figure 6. Dataset after distance calculation.

A snapshot of dataset with Haversine distance between initial and final locations.

Figure 7. Bus trip locations.

Bus trip localization with latitude and longitude.

Figure 8. Bus trip routes.

Complete Bus Trip Routes in the Addis Ababa city.

Table 4. Execution time, R² score, MAE, MSE and RMSE of each Model.

Model Name	Execution Time (In Second)	MAE	MSE	RMSE	R^2 score
Random Forest Regressor	179.254699	0.13677	0.05404	0.23247	0.9995
Gradient Boosting	45.98444605	0.304526	0.149515	0.386671	0.99039
Linear Regressor	6661.479553	0.324486	0.162654	0.403303	0.989363
Lasso Regressor	4257.271728	0.364665	0.182817	0.427571	0.987939
K- Nearest Neighbor	7054.738146	0.469807	0.463268	0.680638	0.979969
Support Vector Regressor	7651.254699	0.136771	0.054043	0.232471	0.999496
Ensemble Averaging	7657.145359	0.494396	0.473161	0.691622	0.999967

Figure 9. Comparison of models’ performances.

Performance Comparison of all the Six Models’ and Ensemble model.

Figure 10. Model testing result.

Testing Result of Random Forest model.

Figure 11. Predicted and actual values of different models.

Predicted and actual values as generated by various models (A) Random Forest (B) Gradient boosting (C) Linear Regressor (D) Lasso Regressor (E) K Nearest Neighbor (F) Support Vector Regressor (G) Ensemble Averaging model.

Table 5. Linear regression analysis of the individual models.

Model	Parameter	Approximated Value	SE	p	t
Random Forest Regressor	A	0.9913	0.005	0	187.342
	B	28.5336	16.121	0.75	1.86
Gradient boosting	A	1.0012	0.06	0.098	176.143
	B	−12.2083	153.165	0.931	−0.08
Linear Regressor	A	0.9844	0.005	0	192.189
	B	44.7213	16.195	0.006	2.762
Lasso regressor	A	0.9099	0.005	0	196.178
	B	104.2845	62.836	0.098	1.66
K Nearest Neighbor	A	0.9876	0.05	0	179.143
	B	−13.1023	149.135	0.9021	−0.08
Support Vector Regressor	A	0.9612	0.005	0	189.189
	B	45.3421	16.005	0.006	2.3001
Ensemble	A	0.9311	0.005	0	183.184
Averaging	B	46.0421	15.005	0.006	2.2001

Figure 12. Variables and their SHAP values of the random forest model.

SHAP analysis results for the Random Forest model.

Figure 13. Bus routes according to their final coordinates.

Bus routes and self-organising map clustering schema.

Figure 14. Maps according to possible cluster of the data for bus routes.

Varied generated Maps according to possible charging station placements.

Figure 15. Possible locations for EV charging stations according to SOM.

Possible locations for EV charging stations as resulted by applying SOM.

Figure 16. Comparative clustering performances.

Comparison of various clustering performances (a) Silhouette score based comparision (b) Number of clusters (c) Silhouette score vs Homogeneity score.

Figure 17. Optimal EV charging point locations in the map of Addis Ababa.

EV charging point locations in Addis Ababa city.

Beshah, T., and S. Hill. 2010. “Mining Road Traffic Accident Data to Improve Safety: Role of Road-Related Factors on Accident Severity in Ethiopia.” In 2010 AAAI Spring symposium series.

 Google Scholar

Gu, X., C. Chen, Y. Yang, X. Miao, and B. Yao. 2021. “Reliability Prediction of Further Transit Service Based on Support Vector Machine.” Measurement and Control 54 (5–6): 845–855.

 Web of Science ®Google Scholar

Liu, D., J. Sun, and S. Wang. “Bustime: Which Is the Right Prediction Model for My Bus Arrival Time?.” In 2020 5th IEEE International Conference on Big Data Analytics (ICBDA), pp. 180–185. IEEE, 2020.

 Google Scholar

Wang, Y. 2021. “Intellectualization of the Urban and Rural Bus: The Arrival Time Prediction Method.” Journal of Intelligent Systems 30 (1): 689–697.

 Web of Science ®Google Scholar

Shanthi, N., V. E. Sathishkumar, K. Upendra Babu, P. Karthikeyan, S. Rajendran, and S. Muhammad Allayear. 2022. “Analysis on the Bus Arrival Time Prediction Model for Human-Centric Services Using Data Mining Techniques.” Computational Intelligence and Neuroscience 2022.

 Web of Science ®Google Scholar

Zhou, X., P. Dong, J. Xing, and P. Sun. 2019. “Learning Dynamic Factors to Improve the Accuracy of Bus Arrival Time Prediction via a Recurrent Neural Network.” Future Internet 11 (12): 247.

 Web of Science ®Google Scholar

Prakash, R., L.-S. Tey, E. Chakravorty, and K. L. Hurley. 2019. “Bus Arrival Time Modeling Based on Auckland Data.” Transportation Research Record 2673 (6): 1–9.

 Web of Science ®Google Scholar

Li, J., J. Gao, Y. Yang, and H. Wei. 2017. “Bus Arrival Time Prediction Based on Mixed Model.” China Communications 14 (5): 38–47.

 Web of Science ®Google Scholar

Treethidtaphat, W., W. Pattara-Atikom, and S. Khaimook. 2017. “Bus Arrival Time Prediction at Any Distance of Bus Route Using Deep Neural Network Model.” In 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC), pp. 988–992. IEEE.

 Google Scholar

Ou, Y. 2022. “AI for Real-Time Bus Travel Time Prediction in Traffic Congestion Management.” Humanity Driven AI: Productivity, Well-Being, Sustainability and Partnership 63–84.

 Google Scholar

Fan, W., and Z. Gurmu. 2015. “Dynamic Travel Time Prediction Models for Buses Using Only GPS Data.” International Journal of Transportation Science and Technology 4 (4): 353–366.

 Google Scholar

Data availability statement

The data that support the findings of this study are available from the corresponding author, T.K.Das, upon reasonable request.