The amount of textual data generated in environments such as social media, blogs, online newspapers, and so on, have attracted the attention of the scientific community in order to automatize and improve several tasks that were manually performed such as sentiment analysis, user profiling, or text categorization, just to mention a few. Fortunately, several of these activities can be posed as a classification problem, i.e., a problem where one is interested in developing a function, from a set of texts with associated labels, capable of predicting a label given an unseen text. In this contribution, we propose a text classifier, named $\mu$TC. $\mu$TC is composed of a number of easy to implement text transformation, text representation and a machine learning algorithm that produce a competitive classifier even over informal written text when these parts are correctly configured. We provide a detailed description of $\mu$TC along with an extensive experimental comparison with the relevant state-of-the-art methods. $\mu$TC was compared on 30 different datasets obtaining the best performance (regarding accuracy) in 18 of them. The different datasets include several problems like topic and polarity classification, spam detection, user profiling and authorship attribution. Furthermore, it is important to comment that our approach allows the usage of the technology even for users without knowledge of machine learning and natural language processing.