Water quality monitoring faces challenges such as the large number of measured parameters, differences in data availability, and missing data. This study proposes an Availability-Aware Feature Selection (AAFS) method for classifying the Water Quality Index (WQI). The method ranks features by combining their predictive importance with their missing-data rate, and determines the number of selected features automatically. It was evaluated on 1,991 records from the public Water Quality Data dataset containing eight water quality parameters, and compared with using the full feature set across eight classification algorithms and 20 independent data splits. Results showed that, under default settings, AAFS identifies BOD as the most important feature, and that the effect of feature selection depends on the classification algorithm. AAFS-selected features increased the accuracy of SVM, KNN, and Naive Bayes by 13, 22, and 20 percentage points, respectively, whereas the full feature set performed better for random forest, gradient boosting, and decision tree, and AdaBoost performed identically in both configurations. Performance differences were statistically significant for seven of the eight algorithms. This pattern can be explained by the concentration of most predictive information in BOD; therefore, feature selection is not always beneficial and should be matched to the algorithm and the intended application. AAFS can be useful for lightweight, low-cost monitoring systems, but for tree-based and ensemble models, retaining the full feature set is preferable. Because the WQI label is computed from the input parameters themselves, generalizing these findings requires evaluation on datasets with independent labels.