Detection of cracks at the earliest stage is crucial, as these are the primary indicators of infrastructure's health. Manual inspection is often carried out for infrastructure inspection which requires in-depth knowledge of domain, which is time-consuming, labor intensive. The in-accessibility of infrastructure in manual inspection make it more challenging and complex. Therefore, various efficient and fast image-based automatic techniques have been introduced in the literature for concrete crack detection task. This paper aims to evaluate the performance six hand-crafted features based traditional approaches in comparison with deep Convolutional Neural Networks (CNN's) for concrete crack detection using different performance metrics. The dataset is obtained by combing data from two publicly available datasets and consists of 40000 crack and non-crack images. Extensive experiments are conducted demonstrating that Random Forest and KNN classifier performs better with 98% accuracy with Area Under the Curve 0.99 as compared to the other classifiers using handcrafted features as well it is faster than deep convolutional neural networks. The computational time for the DCNN is larger than all other classifier but it has the capability to extract feature from images automatically.