A Hybrid CNN-Logistic Regression Model for Accurate and Efficient Car Damage Assessment

This paper develops a machine learning model to classify the various sorts of car damages with high performance toward damage detection. Precision, recall, and F1-score values for all types of damages are tremendously high, ranging between 97.51% and 97.70%. In particular, Bumper Dent, Scratch, Glass Shatter, Door Dent, and Smash have almost identical performances (97.51%, 97.71%, 97.65%, 97.54%, and 97.70% respectively), thereby verifying the dependability of the model in detection as well as categorization of such damages. The dataset utilized to train the model contains 12,630 images, and the overall accuracy achieved by the model is 97.62%. The dataset provided includes a large number of images, and support values for the two highest include Scratch with 2620 images and Smash with 2610 images, respectively, both constituting 21% of the dataset. Support values ranging from 2410 to 2550 images occur for Bumper Dent, Glass Shatter, and Door Dent. Overall model accuracy is 99%, affirming the robustness of the model in making a distinction between types of car damage. These results give evidence for the model's efficacy in automating damage detection, which can also be pivotal in insurance claims and repair assessments. For these, and even more promising, applications in the automotive industry, its effectiveness across the considered categories leads to a complete revolution of car damage detection and the consequent streamlining of processes in the automotive industry.