Most of us know the adage, “Garbage in, garbage out.” AI is only as powerful as the quality of data that fuels it. From collection to deployment, each stage of the AI value chain matters. However, an often overlooked (and highly problematic) area is the middle layer: data preprocessing and feature engineering.
This is what Dr. Joseph Nyangon, a Kenyan inventor and AI expert, tackled head-on over two years, ending up with a game-changing innovation that enhances the reliability of AI models. His work, awarded US Patent 12,190,219 B1, brings us a multi-stage outlier detection system that systematically identifies and removes bad data before it corrupts AI training. Nyangon’s approach effectively filters low-quality data from harming machine learning models by applying quantile-based thresholds and matrix decomposition techniques. And just like that, a persistent bottleneck that has long plagued AI systems is now solved.
Why This Matters: A Fix for AI’s Data Problem
Bad data’s time- and profit-draining cost has long been a challenge for users. In industries like supply chain management, even minor data inconsistencies produce a “butterfly wing effect” of sorts—small perturbations that can eventually cause storms of disrupted AI-driven forecasts, inventory miscalculations, and inflated operational costs. AI models trained on corrupted datasets will almost certainly lead to significant inefficiencies, ranging from excessive stockpiles to missed delivery windows.
Dr. Nyangon’s innovation, solving this issue at the source, helps strengthen the entire AI ecosystem. Expensive computational fixes such as larger datasets or more complex algorithms are avoided, as Dr. Nyangon’s solution optimizes data prior to the start of model training—drastically reducing computing costs and boosting predictive accuracy in real-world applications.
A Global Innovation, Built with SAS Viya
One thing that makes Dr. Joseph Nyangon’s work particularly noteworthy is how he built it. He utilized SAS Viya, a powerful cloud-based AI and analytics platform, to develop and refine what eventually became a patented outlier detection system. The platform’s machine learning and data analytics capabilities were crucial in Joseph’s design, testing, and validation of his multi-stage preprocessing approach.
His work also points to a paradigm shift: Silicon value and most other centers from which AI innovations have flowed to Africa and the rest of the world! We are witnessing high-impact novelties from diverse, global talent pools like Kenya. Nyangon demonstrates how African inventors and technologists actively contribute to AI’s future. We are not just consuming it.
Fixing the “Invisible” AI Problem
Until recently, breaking news about AI has been about bigger models (ChatGPT 4.5, I’m looking at you!), faster processors, or flashy breakthroughs at the extremes of the value chain. Dr. Nyangon’s work proves that some of the biggest wins come from fixing what happens in the middle of the value chain. As users in government and private sectors expand their investments in AI, they’d be wise to look beyond bigger or faster but also at more intelligent, cleaner, efficient data pipelines. Thanks to Joseph’s patent award-winning breakthrough ( Patent 12,190,219 B1), AI value chains will be stronger, more reliable, and ready to deliver real business impact in Africa and elsewhere.
Great invention, thank you @Lawrence for compiling and sharing this game changing innovation