Recently, development of a proactive intelligent e-commerce environment is one of the important research issues. While, depending on the type of the transaction, different techniques could be utilized to improve the e-commerce system and handle multiple scenarios. Here, we present a new proactive system with proposed techniques that have been implemented to interact within an e-commerce environment. Complete description of this intelligent e-commerce system with the proposed techniques is the goal of this thesis. In this thesis, we propose an environment implemented to work in the e-commerce field and with market basket data. A market basket data is a database that consists of a set of transactions. We developed a proactive e-commerce system by focusing on the use of different intelligent techniques to improve the management process of both the products and the customers inside the e-commerce environment. We also present the intelligent prediction idea, and describe the implementation of the fully automated framework, named PIEE (Proactive Intelligent E-Commerce Environment). PIEE framework consists of four main integrated algorithms, in addition to our prediction techniques, those algorithms are the following: (1) F-Tree algorithm: clustering dataset, (2) PWO measurement and Overlap Estimator algorithm: detecting the neighbor threshold values and measuring the similarity, (3) Classifier algorithm: classifying data, and (4) F-Rules algorithm: generates association rules set. Firstly, we have to cluster the transactional dataset to break up the unstructured data. The first step in cluster analysis is to select a measure of similarity, and the key difference between most clustering methods is in defining and choosing the similarity measurement function. So, we present a new solution to solve the overlapping problem between clusters, and propose a new similarity measurement function called PWO “Probability of the Weight Overlapping”. One of our similarity measurement key advantages is that it uses the percentage as the reasonable degree between similar items. In addition to the following: (1) PWO is a global metric, depends on only the total weight of each item in a cluster, rather than the number of transactions in a cluster or the number of items in each transaction, which makes it fast in the measurement process. (2) PWO can determine the best pair of clusters by measuring the similarity degree between different clusters. (3) PWO can use as a classification metric based on the similarity between items. In this thesis, we used this similarity measurement in: (1) Measuring the similarity between clusters of categorical or transactional dataset, and then we use the result of similarity measure function in order to determine the best pair of clusters to be merged. (2) Classify the dataset based on the similarity between categorical items. Secondly, certainly the closeness of clusters changes according to the dataset type and form; then the overlapping between clusters varies depending on the behavior of the dataset, so our proposed framework detects the best similarity value for different datasets according to a small training dataset. Thirdly, clustering data will enable each cluster to express their own associations without interference from other subgroupings that have different patterns of relationships, and the rules produced by each cluster are more informative than rules found from direct association rule mining on the un-partitioned dataset. Also fast and accurate clustering of transactional data has many potential applications in e-commerce intelligence. But there are several challenges when clustering categorical attributes; the most common are: (1) No Natural Order, (2) High Dimensionality, and (3) Subspace Clusters. So, we propose a novel algorithm for clustering transactional data called F-Tree. The simple idea behind the F-Tree is to generate small and highly pure clusters, and then merge them. That makes it fast, and dynamic in clustering large transactional datasets with high dimensions. F-Tree proved to handle the rare items; which is powerful in detecting the rare clusters. On the other hand, the main disadvantage is that it generates a large number of small clusters and most of them have a large similarity. Fourthly, Association rules, in our environment, are an important technique to detect common relationships between items. The most challenging part of association rule inference involves finding sets of items which satisfy specific criteria, and in turn they are used to infer the rules themselves. There are mainly three problems to deal with mining association rules. (1) The algorithmic complexity, where the number of rules grows exponentially with the number of items, (2) Interesting rules must be picked form the set of generated rules, and (3) Generation of rare rules that have lower frequency in each of their individual terms. For this, we developed new algorithm called “F-Rules”, for fast and efficient association rules mining, and we improved both accuracy and efficiency by employing F-Tree data structure, to compactly store and efficiently retrieve a large number of rules. An advantage of F-Rules is the saving in main memory usage, and can be explained from two aspects. F-Rules use F-Tree; the compactness of F-Tree brings significant gain in storing a large set of rules where many items in the rules can be shared, and F-Tree is an index structure of rules. Also pruning the F-Tree at level equals to the minimum support, contributes to the saving of main memory especially when there is a huge number of rules, and long pattern rules. Fifthly, classification plays an important role in supporting business and scientific decision-making. We have examined two major challenges in classification: (1) efficiency at handling huge number of mined association rules, and (2) effectiveness at predicting new class labels with high classification accuracy. We propose two new algorithms for classification transactional data, the first algorithm is based on the F-Tree structure for building a classifier which uses the idea of building F-Tree data, while the second is based on the PWO similarity measurement for building a classifier which uses the idea of closeness to identify the transaction’s class. As the two algorithms’ ideas are different, also the performance, accuracy and features are different. We also present the same two approaches but with a dynamic updating feature to the classifier model which reflects the dynamic changes in dataset direction. Experimental results show that our classifier is more accurate than produced by state-of-art classification system. In this point we make the following contributions: (1) Instead of relying on a single rule for classification, we determine the class label by a set of rules. (2) Handling new cases, by proposing a dynamic classification method with updating features. Finally, we propose a novel method for prediction of products’ percentage, vital products’ effectiveness, recommended products, and customers’ tendencies. We prove that this prediction system directly improves the e-commerce marketing. It is hoped that the proposed environment would illustrate some practical ideas and advantages which can be beneficial for the e-commerce environments.