The challenging problem of classifying big data, exemplified by volume, velocity, variety, value, and veracity, is very relevant in all fields of study. Additionally, the real-time datasets also suffer from another problem of imbalanced class data, which is common among various application-based data such as healthcare, intrusions, fraud detection, stock prediction, weather forecasting, etc. To handle big imbalanced data, this paper presents an advanced parallel programming-based classifier using an optimised Extreme Learning Machine (ELM) classifier. The proposed MapReduce-based Hyper-parameter Optimized ELM (MR-HPOELM) is developed by integrating the MapReduce framework into the classification model along with optimising the ELM configurations by tuning the hyperparameter values using Black Widow Optimization (BWO). The proposed MR-HPOELM also uses a hybrid feature selecting method by combining the Information Gain (IG) and Population and Global Search Improved Squirrel Search Algorithm (PGS-ISSA). This hybrid feature selection approach reduces the feature dimension using information gain and then selects the optimal feature subset using PGS-ISSA. Finally, the MR-HPOELM obtains these selected features and classifies the data accurately, irrespective of the class imbalance problem. Implemented in MATLAB, this hybrid model is tested on benchmark datasets where outcomes demonstrated that for the voluminous class unbalanced datasets, the suggested IG-PGS-ISSA and MR-HPOELM based classification models produced excellent classification accuracies with reduced computations.