HEAT level control (HLC) is one of the important elements for operating an iron-making blast furnace (BF). The goal of HLC is to maintain the hot metal temperature (HMT) as close to a preset aim as possible. HMT is an important indicator of both the product quality and fuel efficiency, and is measured from tapped out liquid iron. For instance, high values of HMT mean unnecessary fuel consumption together with sub-optimal hot metal chemistry, whilst low values of HMT may indicate insufficient fuel consumption, which may consequently lead to dangerous situation of freezing the slag inside the BF. Once an aim HMT is decided, based on production and plant constraints, several inputs of the BF can be adjusted by operators to control this HMT. However, any change of the inputs requires time to take effect, due to inherent time lags, and the best practice control relies heavily on an operator’s experience and judgment. This is due to the complexity of the numerous BF processes, which essentially can be described as highly non-linear, stochastic and nonstationary in nature. The goal of this work is to employ a data mining approach to analyse the BF system and build a dynamical modelling system, which will generate a set of understandable symbolic rules for prediction of HMT changes. The model is regularly generated by decision tree applications, See5 and Cubist, in order to adapt any significant contextual variations of the BF. This rule based system can help the operators increase the BF’s efficiency by making timely control adjustments with the goal of minimising variation of HMT with time. These models also provide a corpus of driving factors that can also be analysed by domain experts as an objective knowledge source.