Our team consists of:
- Muhammad Givari Ramdhani
- Fauzan Ihza Fajar
- Roissyah Fernanda Khoiroh
Aiming to forecast delivery delays within e-commerce transactions, we leveraged the Random Forest, XGBoost, and Extra Trees Classifiers. Utilizing accuracy as the key performance indicator, we meticulously optimized the models and delved into the available features. Ultimately, the Extra Trees Classifier emerged as the most adept for delay classification, achieving an impressive 93.879% accuracy. This exceptional performance propelled our team to a remarkable 7th place out of 70 competitors, securing a coveted spot in the semifinal round.
The top 20 semifinalists were tasked with harnessing web scraping and Natural Language Processing (NLP) to gain deeper insights into customer sentiment. Our team employed Sentiment Analysis and Topic Modeling to unravel the underlying sentiment and themes within customer reviews of a prominent delivery expedition service. The data was scraped from Twitter and the Google Play Store.
The dataset comprised 13,952 rows, with a distribution of 25.7% positive, 40.2% negative, and 34.1% neutral reviews. LDA identified one optimal topic for neutral sentiment, sixteen for positive, and three for negative. Interpretation revealed that positive reviews commended the service's quality, advocated for it, and highlighted the convenience of its digital offerings. Conversely, negative reviews primarily decried delivery delays and service shortcomings. Neutral reviews, meanwhile, centered around updated receipt tracking information. In the final round, we have to present our work. We were elated to secure third place.