The Effect of Initialization Weights for Multi Layer Perceptron Performance on Prediction of House Construction Costs

Abstract

The house is one of the primary human needs besides food and clothing. Therefore, the community will always try their hardest to meet primary needs. For the middle and lower class people, it is going to very difficult to build a residential house because the income does not match the increase in house prices. With an artificial neural network, the middle to lower class people can estimate the costs that must be prepared if you want to build a residential house, of course this will be cheaper than using housing developer services. Based on the data that has been obtained, the researcher is then trained and tested using an artificial neural network with 13 data input, 25 hidden layers, a learning rate of 0.75, the number of iterations of 1000, the best test results are MSE value, 0.1, mean accuracy of 97.22 and computation time of 0,028 seconds