Abstract— In this paper, a review of prediction techniques suitable for ambient intelligence environments is presented. Prediction challenges in sensor networks are considered in two phases including pattern extraction and rule matching. The prediction techniques reviewed in this paper come from two main research areas, namely, data mining and soft computing techniques. Moreover, a statistical modelling technique based on Markov chain is also considered. In this paper, we identify the centralized and distributed techniques of both data mining and soft computing areas. In addition, we identify the distributed approaches that utilize computational power of sensors in an ambient intelligence environment. Moreover, we show that some techniques use compression, regression or fuzzy methods to reduce the size of the collected sensory data. I. INTRODUCTION With growing interest in the use of smart environments, a new generation of such environments has drawn attentions of many researchers [14]. The predictive environment as the third generation of smart environments will provide more intelligence capabilities in comparison with its former generations [15]. The predictive environment is an ambient intelligence environment that facilitates the interactions of occupants, devices and environment. Energy saving, convenience of occupants, safety and security are the main objectives of smart environments. Despite the lack or minimum usage of sensors, the first and second generations of smart environments had some success to meet some of the above objectives. For instance, a bank equipped with an old security system in which employees can press a button to call the police in a bank robbery, is one of the first generation of smart environments. Second- generation smart environments are equipped with individual sensors or a sensor network. This generation of smart environments is also called an automatic environment. A smart building with automatic lights and heater control is an example of the second generation. New emerging third generation of smart environments also known as predictive environments have both manual and automatic control features; moreover, it can learn from environmental changes as well as behavioural patterns of occupants. Predictive M.Javad. Akhlaghinia is with the School of Computing and Informatics, Nottingham Trent University, Clifton Campus, Nottingham, NG11 8NS, UK (phone: +441158488375; email: mohammadjavad.akhlaghinia@ntu.ac.uk ). Ahmad Lotfi is with the School of Computing and Informatics, Nottingham Trent University, Clifton Campus, Nottingham, NG11 8NS, UK (phone: +44 115 8488390; email: ahmad.lotfi@ntu.ac.uk ). Caroline Langensiepen is with the School of Computing and Informatics, Nottingham Trent University, Clifton Campus, Nottingham, NG11 8NS, UK (phone: +44 115 8488367; email: caroline.langensiepen@ntu.ac.uk ). environment collects data acquired from a sensor network. Collected data include variety of attributes, such as environmental changes and occupants interactions with environment. These data are used in a learning approach to make a predictive environment that can predict the occupancy of different areas as well as requirements and interests of occupants at different times. This predictive feature steps up the performance of energy saving approaches in a smart environment; in addition, it improves the convenience of occupants as well as security and safety. Data collection and prediction are two challenges of predictive environments. The first challenge is due to the energy and bandwidth constraints in sensor network [12-13], but the second challenge which is the main focus of this article is a learning challenge in distributed sensor networks. Predictive environment can predict the next state of consecutive interactions with the use of the knowledge it has learnt from previously observed interactions. For instance, it can predict favourite light intensity of different occupants in a specific area of the environment at a specific time of a day. Prediction consists of first pattern extraction to identify sequences of actions, and then sequence matching to predict the next action in one of these sequences [1]. In this paper, a comprehensive review of available techniques suitable for the third generation of ambient intelligence environments is presented. Section II is an introduction to different prediction techniques. Section III is a review of data mining techniques and section IV is a review of soft computing techniques. A statistical model for prediction in predictive environments is considered in section V. Relevant conclusions and future works are drawn in the last section. Note that this is an ongoing research project, the ultimate goal of which is to build a predictive ambient environment. II. PREDICTION TECHNIQUES IN PREDICTIVE ENVIRONMENTS Prediction in predictive environments is a learning challenge in distributed sensor networks. There are variety of techniques for this learning challenge including data mining techniques and soft computing techniques. Some of these techniques are centralized approaches such as some data mining techniques and some of them are distributed approaches such as agent-based techniques. In addition, some of these approaches apply compression or fuzzy methods to reduce the amount of stored sensory data. Soft Computing Prediction Techniques in Ambient Intelligence Environments M. Javad Akhlaghinia, Ahmad Lotfi, Member, IEEE and Caroline Langensiepen 1-4244-1210-2/07/$25.00 ©2007 IEEE. 1614