Predictive analytics has multiple facets that range from failure predictability and optimal asset management to high-level managerial insights. Predicting the failure time of assets and estimating their reliability through efficient prognostics and reliability assessment framework allow for appropriate maintenance actions to avoid catastrophic failures and reduce maintenance costs. Most of the systems used in the manufacturing and service sectors are composed of multiple interdependent components. Moreover, these systems experience dynamic operating conditions during their life. The dynamic operating conditions and the system complexity pose three challenging questions: how to perform the prognostic and reliability assessment of a complex multi-component system, how to perform the prognostic and reliability assessment of a system functioning under dynamic operating conditions, and how to use the condition based and reliability assessment data to find the optimal maintenance strategy for complex systems. This dissertation investigates five tasks to address these challenges. (1) To capture the stochastic dependency between interdependent components of a system through a continuous time Markov process with the transition rate depending on the state of all components of the system. This technique helps get an accurate estimation of system reliability. (2) To propose a framework based on instance-based learning to predict the remaining useful life (RUL) of a complex system. This technique can be used for highly complex systems with no need of having prior expertise on the system behavior. (3) To incorporate time-varying operating conditions in the prognostics framework through a proportional hazards model with external covariates dependent on the operating condition and internal covariates dependent on the degradation state of the system. (4) To propose a prognostic framework based on deep learning to predict the RUL of a system working in dynamic operating conditions. This framework has two main steps: first identifying the degrading point and developing the Long Short-Term Memory model to predict the RUL. (5) To propose an efficient algorithm for reliability analysis of a phased-mission system, its behavior changes at different phases during the mission. This technique accounts for imperfect fault coverage for the components to get accurate reliability analysis.