Cervical cancer (ca cervix) is one of most common cancer in women and will be highly treatable because of screening. It implied that ca cervix should be diagnosed as early as possible. Therefore, the prediction model is needed to support the early detection and screening of ca cervix. This present study is aimed to make and evaluate predictive model of ca cervix by machine learning approach using open imbalance data. The machine learning approach was implemented by four steps, consists of data pre-processing, feature selection, predictive model development, and model performance evaluation. We used features in Waikato Environment for Knowledge Analysis (WEKA) version 3.8.4 for all of those steps, including imputation, interquartile range, Synthetic Minority Oversampling Technique (SMOTE), resample, and experimenter. The result of this study showed that all of predictive models (RandomForest, Bagging, LogitBoost, ClassificationViaRegression (CVR), and RandomCommittee) were acceptable for ca cervix prediction. The highest accuracy of all predictive model was 99.34%. Our proposed predictive model with highest performance is LogitBoost. The performance of ca cervix prediction using imbalance data will improve if we can handle the missing value, outlier, and data imbalance. We concluded that our proposed predictive model showed the good performance and may be visible for noninvasive ca cervix screening.