Hi,
To categorize the multiple states of the fan(ON, OFF, CLOG, FRICTION), I am using Application Software Pack - ML-based System State Monitor on LPCXpresso55S69.
On the validation dataset, the prediction accuracy percentage for the confusion matrix for all states is greater than 95%. The model, however, is unable to distinguish between the ON and CLOG states with great clarity when tested on a real-time dataset. The outcome alternates between two states. I saw in the Jupyter notebook that the model was trained by only using Ax, Ay, and Az data. I modified the input dataset in the Jupyter-notebook to add more data by integrating Bx, By, and Bz data and deployed the model on the MCUXpresso IDE to obtain more accurate results on swinging states. Real-time test results showed an even worse outcome. All four states are not recognized.
What changes should be made in the IDE's source code so that it correctly interacts with the new model and processes real-time data?
Thank you so much!
#LPCXpresso55s69 #ml_state_monitor_cm33 @https://community.nxp.com/t5/eIQ-Machine-Learning-Software/Application-Software-Pack-ML-based-System-State-Monitor/ta-p/1413290
