A material's electronic properties and technological utility depend on its band gap value and the nature of band gap (i.e. direct or indirect). This nature of band gaps is notoriously difficult to compute from first principles. In fact it is computationally intense to approximate and also rather time consuming. Hence its prediction represents a challenging problem. Machine learning based approach offers a promising and computationally efficient means to address this problem. Here we predict the nature of band gap for perovskite oxides (ABO3) with elemental composition, ionic radius, ionic character and electronegativity. We do this by training machine learning models on computationally generated datasets. Knowing the nature of the band gap of the perovskite oxides (whether direct or indirect) plays a pivotal role in determining whether the perovskite can be used for photovoltaic or photocatalytic applications. A total of 5329 perovskite oxides are considered in this study. Here, we determine the correlation between the nature of band gap and the composition of the perovskite oxide. A Random Forest algorithm is used for predicting the same since it yielded higher accuracy (∼91%) compared to the other Machine Learning models. The approach suggested here can be used to predict the nature of bandgap and can also aid in novel materials discovery within the family of perovskites. This is a robust, quick, and low-cost strategy to find novel materials for light harvesting applications in particular. Also we present feature ranking as it pertains to prediction of nature of bandgap and also discuss correlation between the features. We also show feature importance graphs and SHapley Additive exPlanations (SHAP) as is relevant for prediction of nature of band gaps. Using the approach reported, NaPuO3 and VPbO3 are discovered to be good candidates for solar cell materials (direct band gap∼1.5 eV). Novel composition predictions for targeted applications are the future and our model is a step ahead in this direction.