Sepsis is a condition which needs critical care and timely intervention for improved patient outcomes. A detailed data preprocessing pipeline was implemented, which consists of KNN imputation for handling missing values and IQR-based filtering to eliminate outliers. ADASYS (Adaptive Synthetic Sampling) was used to address the class imbalance issue. A Multi-Layer Perceptron (MLP) classifier was optimized through grid search,the parameters such as solver, learning rate, and momentum are considered for refining. The optimized model achieved an accuracy of 97.95%, a precision of 0.5714, and an F1 score of 0.0327, with a ROC-AUC of 0.7973, demonstrating strong discriminatory power in predicting sepsis. This research highlights the importance of meticulous data preprocessing, class balancing, and hyperparameter tuning in developing machine learning models for critical healthcare applications. The findings contribute to advancing the development of predictive systems for early sepsis detection in clinical environments.