Interpretable machine learning methods help users understand model decisions when transparency and trust are important aspects of the model. A widely used method is Partial Dependence Plots (PDP), which assumes feature independence, leading to misleading interpretations. To address this limitation, we propose an improved version of PDP, termed iPDP, which incorporates a distance-based filtering technique to remove data points that are distant from the reference instance. This results in more accurate and nuanced interpretations of feature effects. We evaluate iPDP using the cervical cancer dataset. Our experimental results show that iPDP better captures feature interactions and provides a more reliable understanding of model behavior, particularly when features exhibit correlations. The choice of distance and threshold parameters influences the quality of the interpretability, with L2 distance generally providing smoother and more consistent results. This work demonstrates the potential of iPDP as a valuable tool for enhancing interpretability in complex machine learning models.