Predicting genotype-phenotype relationships is essential for identifying disease-associated SNPs, predicting disease risk, early disease detection, and case-control classification for screening. This study evaluates the performance of a range of machine learning and deep learning algorithms, along with polygenic risk score (PRS) tools, using 80 binary phenotypes from the openSNP dataset. We extracted genotype data for each phenotype, conducted phenotype transformation, and subjected the genotype data to initial quality control procedures using PLINK. We partitioned the genotype data into base and target sets, created the Genome-Wide Association Studies summary statistic (GWAS) file from the base data, and provided both the GWAS file and target data to the PRS tools for prediction. In the case of machine and deep learning algorithms, we applied p-value thresholding to the training data to select the single nucleotide polymorphisms (SNPs) highly associated with the phenotype in question and used this refined data for prediction. For each phenotype, we reported the average 5-fold AUC for the best-performing model among 29 machine learning algorithms, 80 variants of deep learning algorithms resulting from different combinations of hyperparameters, and three polygenic risk score tools with 675 combinations of clumping and pruning parameters. Machine learning demonstrated better performance for 44 phenotypes, while polygenic risk score tools were more effective for 36 phenotypes. It was observed that, in most cases, the XGBoost (machine learning), ANN (deep learning), and Plink (PRS) tools consistently delivered the best outcomes. The effectiveness of machine/deep learning algorithms and polygenic risk scores in predicting phenotypes is influenced by several factors, including hyperparameters, classification methods, the specific phenotype under consideration, data quality, and data quantity.