Implementation of HMLP network with different activation function for cervical cells classification
Saiful Zaimy Yahaya, Nor Ashidi Mat Isa · 2011
In this paper, the implementation of Hybrid Multilayered Perceptron (HMLP) network with two different activation functions has been studied. The study was conducted for PC based system and hardware based system and was evaluated in classifying the cervical cells data into it classes. For the hardware based system, this system was implemented in the 8051 microcontroller system. The structure of the HMLP network was arranged in two stages of hierarchy for acquiring four (4) types of input data and producing two outputs for each stage of hierarchy. The number of data used for training and testing of the HMLP network was 200 and 120 respectively. The HMLP network was trained using a Modified Recursive Prediction Error (MRPE) algorithm to obtain the appropriate parameters for the network. The result was evaluated in term of the percentage of accuracy, sensitivity, specificity, false negative and false positive. The result showed good performance of HMLP network for both types of activation function for the implementation on PC based system and hardware based system. For single number of hidden node, the performance of HMLP implemented with sigmoid and Elliot activation function was proven to be equals.