RISTEK DATATHON 2023

Our team consists of:

• Fauzan Ihza Fajar
• Mohamad Faza Fauzan
• Roissyah Fernanda Khoiroh

Preliminary Stage

We aimed to predict average vehicle speed for a specific time interval using Random Forest and XGBoost Regressors. We employed SMAPE as the evaluation metric and painstakingly fine-tuned the model and explored available features. Our team achieved an impressive 11th place out of 88 competing teams, securing a spot in the final round. We were further recognized with awards for top Data Preprocessing and top 3 Exploratory Data Analysis.

Final Round

The top 10 finalists were tasked with building a system or developing an algorithm to address traffic congestion, with local deployment as the key objective. Our team leveraged clustering, regression, and classification algorithms to forecast congestion levels at ten intersections equipped with AI-powered ITCS. We found KMeans clustering with a silhouette score of 0.596 to be effective in categorizing travel times into four groups. Furthermore, XGBoost Regressor Tuning yielded the lowest SMAPE score at 1.714, while XGBoost Classifier Tuning delivered the best performance in congestion level prediction with a precision score of 95.66%. Although we did not secure the top spot, finishing in 4th place proved to be an invaluable experience with significant learning takeaways.