This study proposes a new framework for predicting retail customers' credit risk in LendTech platforms using alternative data and machine learning techniques. Traditional credit scoring methods often fail to capture complex behavioral patterns and perform poorly in digital lending environments. The research utilizes alternative data sources such as digital behavior, transaction history, mobile data, and user interactions to improve default prediction accuracy. Several machine learning models, including Logistic Regression, Random Forest, Decision Trees, and Gradient Boosting, are evaluated. To ensure transparency and interpretability, the SHAP explainability method is applied to analyze feature importance. Results show that ensemble models, especially Gradient Boosting and Random Forest, achieve the highest predictive performance, while behavioral indicators such as transaction patterns, repayment behavior, income stability, and platform interactions are the most influential factors in credit risk assessment. The findings demonstrate that combining alternative data, machine learning, and explainable AI can provide a reliable and effective credit-scoring framework for LendTech ecosystems.