It
is crucial to classify and optimize the thermal behaviors of
isoperibolic batch reactors. This article aims to construct an improved
supervised machine learning-based (ML-based) model to achieve the
task. First, using the minimum redundancy maximum relevance algorithm,
the k-nearest neighbor (k-NN) algorithm
with a feature subset consisting of 18 dimensions is selected due
to the highest total recognition accuracy in 27 different ML algorithms.
Additionally, a cost-sensitive learning approach and Bayesian optimization
algorithm are employed to further optimize the hyper-parameters of
the k-NN model. The accuracies of all data sets using
the optimal k-NN model are all 99.8%, indicating
that the optimal k-NN model has a superior performance
and a good generalization ability. Then, two cases coupled with interpretability
techniques are used to interpret the optimal k-NN
model. Finally, based on the optimal k-NN model,
two novel optimization frameworks (single-objective and multiobjective)
are proposed to optimize the mentioned pilot-scale case, and the results
prove that the optimization frameworks are reasonable and reliable.