Advanced Persistent Threats (APTs) pose a serious challenge to cybersecurity due to their stealthy and prolonged nature. Detecting these attacks early is critical, yet traditional methods often fail to catch subtle patterns or require extensive resources. This study investigates the effectiveness of several machine learning algorithms including Logistic Regression, Random Forest, Naive Bayes, and K-Nearest Neighbors (KNN) to identify APT behavior from system and network activity data. A significant focus was placed on reducing the complexity of the detection system without compromising accuracy. Dimensionality reduction using Linear Discriminant Analysis (LDA) was applied to transform high-dimensional data into a more manageable format, and data cleansing techniques were used to handle null and infinite values. The results showed that the KNN algorithm not only achieved the highest detection accuracy but also maintained performance even when trained on just 10% of the dataset. This approach offers a lightweight, high-precision method for APT detection that can minimize processing overhead while improving real-time responsiveness helping organizations enhance their security posture without sacrificing efficiency.