Cybersecurity is an evolving landscape and the need to discern malicious and benign applications has become paramount. One method of handling this challenge is by deploying machine learning models such as Multi-Layer Perceptron. A critical aspect of working with this model is the choice of activation functions in the hidden layers of the Multi-Layer Perceptron. The activation functions include the identity, logistic, hyperbolic tangent, rectified linear unit, and leaky rectified linear unit. This study aims to discern whether a given app is benign or malicious, primarily by scrutinizing the permissions and intents it requests during execution. Subsequently, a comprehensive analysis is conducted by combining the activation functions in various ways by considering different combinations for the hidden layers. The experimental results show that Leaky Rectified Linear Unit activation provides the highest accuracy (98.22%) when applied as a common activation function across all hidden layers of the Multi-Layer Perceptron. Additionally, in cases involving varied activation functions, the combination of leaky rectified linear unit, rectified linear unit, and hyperbolic tangent in different hidden layers achieves the best accuracy (97.91%) for detecting malicious applications.