Two serious problems affecting the implementation of Human Activity Recognition (HAR) algorithms have been acknowledged. The first one corresponds to non-informative sequence features. The second is the class imbalance in the training data due to the fact that people do not spend the same amount of time on the different activities. To address these issues, we propose a new scheme based on a combination of Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and the modified Weighted Support Vector Machines (mWSVM). First we added the most significant principal components to the set of features extracted using LDA. This work shown that a suitable sequence feature set combined with the modified WSVM based our criterion classifier achieve good improvement and efficiency over the traditional used methods.