Because of their effectiveness in identifying geologic features, seismic attributes are usually used as input to machine learning (ML) models for facies classification. Typically, too many attributes are computed for facies classification, making a predictive modeling task more challenging. Principal component analysis (PCA), a popular dimensionality reduction tool, is widely applied in unsupervised learning. This study investigates the use of PCA to reduce the number of attributes before supervised learning. Our motivation is to maximize the use of seismic attributes, data analytics, and ML to effectively classify the geomorphologic seismic facies of the I-X reservoir of A Field, Offshore Malaysia. A systematic approach is presented, including attribute extraction, dimensionality reduction, feature selection, performance measure, and prediction ability for different classes. We extract 31 attributes that belong to amplitude, Gray-Level Co-Occurrence Matrix (GLCM), instantaneous, geometric, and spectral families. PCA analysis is then carried out to reduce the attribute set of each group into fewer principal components (PCs). We label three classes that combine and represent all the seismic/lithologic facies in the interval. Correlation coefficients, including Pearson, Rank, and Mutual Information (MI) that map the relationship between the input features and the 3-classes output, are calculated to select the optimal subset of features. We train and test support vector machine (SVM), random forest (RF), and neural network (NN) algorithms that are widely used in seismic facies classification. Among the computed seismic attributes, we find that the amplitude, Gray-Level Co-Occurrence Matrix (GLCM), and the spectral group of attributes are the best predictors for the fluvial seismic facies. The trained ML models perform slightly better with seismic attributes than PCs, and only minor differences are observed in the classification results. We find that attribute-to-attribute crossplots and correlation heatmaps effectively facilitate feature selection by improving our understanding of the data redundancy and relevance. © 2022 Elsevier B.V.