<p>In many fields, the availability of appropriate and verified data is a major concern when applying machine learning techniques. The performance of an algorithm is adversely affected by the dimensionality of the data. The objective of this research is to reduce the dimensionality of the data by introducing the GA-based dimensionality reduction technique (GbDRT) and increase the classification accuracy of the machine learning models. The economy is negatively affected by financial fraud. The prediction of financial statement fraud is a crucial problem that needs to be improved. Due to its demonstrated effectiveness, machine learning (ML) is frequently used in the creation of fraud classification systems to detect emerging fraud types. In this study, a novel fitness function and parameter tuning are presented for GA-based dimensionality reduction. The average accuracy and non-correlation values were improved over the GA generations. Thus, the fitness function formula itself is novel, which calculates the fitness value of each chromosome. Three different rates—&#xa0;0.25, 0.5 and 0.75—were examined to identify the ideal crossover and mutation rates. In this study, an improved GA-based dimensionality reduction technique was developed and evaluated on two distinct datasets. Furthermore, a comparison with the standard dimensionality reduction methods was performed. GbDRT algorithm increases the accuracy of the findings, with a maximum accuracy of 99.6% using the Indian dataset and 97.81% using the Taiwan dataset in the ExtraTeesClassifier model.</p>