How Accurate is Kavalytics™? Lets look at KavaLytics V4.2.0 Vs. HPLC results on a recent Hawaiian sample set
One of the most common questions we get is, "How accurate is the system at predicting the potency of Kava?". We decided that instead of rehashing statistics such as MAE, RMSECV, SEP Values, that we'd simply demonstrate on real world samples. This will be the first of many demonstrations of the real world results of KavaLytics™ predictive models on commercially available Kava.
Unsurprisingly, the KavaLytics™ V4.2.0 model's "Noble - Non-Noble" classifier also correctly classified all samples as Noble with 100% confidence. These classifications were verified by their HPLC chemotypes of both Kavalactones and Flavokavains as well as various ratios such as K/DHM and K/FKB. The V.4.2.0 Noble - Non Noble Classifier had an accuracy of 95.1% in predicting Nobility across 500 training samples, including 174 verified Non-Noble samples.
Model V4.2.0 was not trained on these samples and had never seen them before, yet predicted their total Kavalactone contents with HPLC accuracy. These samples, along with many others will be inputted into the new model's (V4.3.0) training set to make the system even more accurate and robust. These days, there are very few types of samples that the model hasn't already seen. Therefore, the majority of new training sample sets include rare cultivars that aren't generally commercially offered, and varieties that appear to have noble Kavalactone chemotypes via HPLC, but that the model correctly predicts as Non-Noble because of their high Flavokavain levels. This will be another blog post for another time.
Thanks for reading!