Herbal medicine has gained its popularity throughout these past few decades as many people start to realize its effectiveness in healing ailments. Many modern and systematic methods have been developed in order to find a potential drug out of herbs. One of them is the Lipinski’s Rule of Five. This project proposes an intelligent microcontroller based potential drug detection system for herbal medicine. The development of the intelligent system employs the investigation of suitable artificial neural network. The input parameters to the neural network are the features of an herb’s molecule as described by Lipinski’s Rule of Five. They are LogP, molecular weights, number of hydrogen bond acceptors and number of hydrogen bond donors. Three types of neural network, namely Multilayer Perceptron (MLP), Radial Basis Function (RBF) and Hybrid Multilayer Perceptron (HMLP), were compared for their performances. The HMLP network trained by Modified Recursive Prediction Error (MRPE) algorithm proved to be the best network which has the highest accuracy of 95%. The obtained optimum HMLP network was implemented into hardware form by using microcontroller. The hardware system yielded a satisfying accuracy of 95%. This project has successfully showed that the system built has high capability in classifying an herb into drug-like or non drug-like categories.
i