Improved AdaboostM1 for Stock Price Prediction Using Multi-layer
Perceptron to Integrate Weak Learners
Abstract
Investment in the stock market is currently very popular due to its
economic gain. Therefore, numerous researchers and academicians work is
focused on financial time series prediction due to its data availability
and profitability. Based on the literature it can be seen that various
versions of the AdaboostM1 algorithm have been applied in the stock
market either by tuning the algorithm parameters or attempting various
base learners but the accuracy has not yet reached to favorable and
reliable level. Therefore, this study proposes an improved version of
AdaboostM1(ADA), which is implemented in the Waikato Environment for
Knowledge Analysis(WEKA) to predict stock market prices based on
historical data. The improved AdaBoostM1 integrates the set of
Multilayer Perceptron (MLP) predictors instead of using DecisionStumps,
which is normally being applied. The enhanced AdaBoostM1 is named
Adaboost with Multilayer Perceptron (ADA-MLP). As the result, the
ADA-MLP was found to be outperforming the original ADA by 1.52%, in
which the ADA-MLP achieved the CA of 100% on average while the ADA
achieved 98.48%. Furthermore, the ADA-MLP was al