This paper presents a comprehensive evaluation of predictive modeling techniques within the field of data science, focusing on the performance of five algorithms: Those are the commonly used algorithms namely Support Vector Machines (SVMs), Logistic Regression, Decision Trees, Neural Networks, and Random Forests. These models were analyzed on a range of datasets: a normal baseline dataset, dataset with 5% data replication, 5% outlier data, 10% data replication, and 10% outlier data. Through the evaluation metrics including confusion matrix, F1 measures, support, accuracy, and a full classification report we see how each model performed and how stable or brittle it was when called on to process data in one or the other way. These findings presented through graphs and charts and supported by the tables lay out a profile of data quality characteristics that affect the performance of the models in interaction with repetitions and outliers. The result provides useful advice to data scientists for model selection and model tuning when facing real-world data quality issues and helps to enhance the decision-making on predictive modeling applications.