Stock prediction using machine learning is an interesting topic for investors. However, the performance of the prediction depends on different techniques and the data itself. In this p aper, a feature selection methodology has been proposed. It co nsists of filter method and wrapper method . A feature selection experiment was conducted on 50 stocks ( SET50) from the Stock Exchange of Thailand (SET). The calculation of feature importance for feature selection was discussed. The feature importance shows how the cohort indicators behave in each wrapping level. Preliminary experiment was conducted to investigate some technical indicators that could be affected by SET50 . The basic machine learning models both regression models and classification models were examined to evaluate the performance of the models based on these features. The proposed f eature selection methodology was flexible and practical as each stock can be influenced by different features. Based on the measured feature importance, the features can be selected in different ways which can efficiently increase the performance of the machine learning model.