Stellenbosch University’s (SU) official business incubator, LaunchLab, held its annual Data Science Hackathon this month, welcoming over 50 students across various faculties to compete for the coveted first place. The event, now in its sixth year, challenges students to tackle a real-world problem statement relating to data science.

The challenge this year was put forward by SU alumni-founded company, and former LaunchLab company, Spatialedge. They challenged students to develop algorithms for anomaly detection in complex cloud billing datasets, a critical issue for businesses managing cloud infrastructure costs.

Securing first place, Team Apollo developed a solution that processes real-time transactional cloud cost data daily, distinguishing between numerical and categorical features for anomaly detection. Using an ensemble of models such as ARIMA and Facebook Prophet, it forecasts expected cost ranges and flags deviations as anomalies. Categorical data, like service type and region, is checked for unusual patterns, identifying outliers such as new or unexpected services. An aggregated daily report, assessing the likelihood of anomalies, is then sent to the customer via WhatsApp, allowing for swift corrective actions to prevent unnecessary expenses.

First place winners:

• Divan van der Bank
• Kharendwe Makhani
• Retief Louw
• Kabelo Maphutha
• Dgibrilly Mutabazi
• Emma Sharrutt

In second place Chekhov’s Puzzle focused on creating a sophisticated cloud cost anomaly detection system designed to provide businesses with actionable insights into their cloud spending. The final product integrated advanced machine learning techniques, user-friendly web design, and a robust backend infrastructure to offer a comprehensive solution. The final product was a scalable tool that would allow businesses to detect, analyse, and respond to cloud cost anomalies in real time.

Second place winners:

• Christopher Johannes Koen
• Hanru Visser
• Chiedza Vushe
• Christiaan Portwig
• Kefas Manda
• Michael Shepstone

Team NeuralRoad secured third place by developing an innovative GCP Billing Analysis Dashboard to help businesses better understand and manage their cloud costs. The team utilized advanced data processing techniques and implemented two key models:

• An ARIMA (AutoRegressive Integrated Moving Average) model for cost forecasting, and a spike detection model using the Z-score method. The ARIMA model analyzes historical monthly cost data to generate a 6-month cost forecast, enabling proactive budget planning with confidence intervals for predictions.
• The spike detection model identifies unusual cost increases on both monthly and daily levels, facilitating quick responses to cost anomalies. The solution features an interactive, four-tab dashboard interface that offers dynamic visualizations, detailed cost analysis tables, and color-coded alerts for easy interpretation. With a user-friendly design and comprehensive approach to cost management, NeuralRoad’s solution stands out for its ability to provide businesses with actionable insights and predictive capabilities in managing their GCP billing.

Third place winners:

• Asisipho Mlobeli
• Desmond Mogodi
• Manda Chuma
• Sthembhiso Rikhotso
• Tando Mahlati
• Ntseliseng Maokola

The judging panel included two representatives from the event sponsor, Spatialedge, Jacques du Toit and Frank Ortman. The Director of our Tech Transfer Office, Ravini Moodley, Killian Boshoff of Sentinell, a LaunchLab resident and the Director of the School of Data Science at Stellenbosch University, Kanshukan Rajaratnam.

The Hackathon is part of a larger initiative by SU LaunchLab to support and encourage entrepreneurship within the university. They have a number of bootcamps and programmes aimed at students, some of which are tailored for specific faculties. In addition to supporting student entrepreneurs, SU LaunchLab incubates the university’s spin-out companies in partnership with the Technology Transfer Office.