This study investigated the capability of machine learning to analyze and predict individuals' motives within an experimental game context. Although humans predict the motives of others to respond appropriately, these motives often overlap and are difficult to tease apart in social exchange contexts. An act of reciprocated favor, for example, could equally be motivated by parochial altruism as by self-interest, and human attributions of motives are notoriously biased. Can machine learning effectively predict motives and offer insights into how individuals prioritize overlapping motives? We analyzed motives in an experimental social exchange game using a Multilayer Perceptron (MLP), Decision Tree (DT), Random Forest (RF), Gaussian Naïve Bayes (GNB), K-Nearest Neighbor (KNN), Support Vector Classifier (SVC), Logistic Regression (LR), and a novel combination of Clustering and Hidden Markov Model (C-HMM). The accuracy, precision, recall, and f1-score were compared in two phases: Phase 1, where individuals focus on a single motive, and Phase 2, where individuals consider multiple motives when making decisions in social exchange. With accuracies of 86.96%, 67.31%, and 70.74% for each class of motives tested in Phase 1, C-HMM outperformed the other models. LR demonstrated the best performance, with an accuracy of 45.57% in Phase 2. Further analysis shows that the strength of relationships with ingroup members is a reliable predictor of reciprocation motives. Our model can be extended to nudging prosocial behavior in human-agent collaborations.