In this paper, we have proposed a mechanism for predicting divorce and utilized the established way to evaluating the scale named Divorce Predictors Scale (DPS). The DPS, which is based on Gottman couple's therapy, consists 54 items self-report questionaries, which could be utilized as features or attributes in a machine learning model. Besides “Divorce Predictors Scale” another form was used that is “personal information form” to collect personal information of participants for a more conventional and disciplined way to conduct this research. Among 200 participants (N=200), 126 (63%) were married and 74 (37%) were divorced. To investigate the success of the Divorce Predictors Scale, Multilayer Perceptron (MLP), Naïve Bayes (NB) and Random Forest (RF) algorithms were used. Using the feature selection approach, we tried to narrow down the significant or most important features based on the correlation-based feature selection. Hence, found 6 features/items influencing the divorce prediction in the context of Bangladeshi data. During applying different algorithms directly on the divorce prediction dataset, the highest prediction accuracy rate was 87.14% with Naïve Bayes algorithm. But after applying feature selection criteria, using the selected features the highest prediction accuracy obtained 84.29% with the Multilayer Perceptron. As indicated by the outcomes, DPS can predict divorce. Family advocates and family specialists can utilize this scale for adding to the arrangement of case definition and mediation plans. Moreover, the analysis could be stated that the affirmation of the Gottman couple's treatment in the Bangladeshi sample are confirmed.