All the cool kids like Bosch and ST Microeletronics are already integrating machine learning and on-chip-training(they call it 'self-learning') into their products. (BMI360/BMI380 from Bosch and NanoEdgeAI from ST Microeletronics) Unfortunately the algorithms are hidden behind seven locks.
So, I decided to write my own little Arduino library for on-chip training. The library is called TinyDecisionTreeClassifier and it's interface mimics DecisionTreeClassifier from sci-kit learn.
What can it do? Well, it helps to automatically classify the data (after the model was trained). I tested the activity recognition (sit/walk/stand), but it would be quite interesting to check, if it can be used for classifying machinery failure based on vibration or for human stress detection from PPG (As in smartwaches). My guess is that it would work. The library is now available from PlatformIO registry or from my github under Apache-2.0 license ('do whatever you want' license).