Plant stress phenotyping based on image-analysis techniques can facilitate targeted interventions and optimize management practices. However, considering abiotic stresses – especially under the influence of climate changes – various technical challenges still need to be overcome, and there is a need to identify functional traits that can be used as markers to monitor plants health status accurately and rapidly. This study aimed to develop a machine learning model able to detect salt stress status in Phaseolus vulgaris, by combining image-based approaches and minimal morpho-physiological/biochemical analyses. Therefore, a flexible pipeline was created, starting with the data acquisition and analysis (feature extraction) followed by the feature selection phase, which involved statistical approaches such as principal component analysis and correlation analysis. Finally, two predictive models based on decision trees were developed to detect stress presence (2-class model) and intensity (3-class model). Specific biochemical parameters and image-analysis-derived traits, like malondialdehyde content, chroma difference and chroma ratio indexes, resulted as key features in categorizing both non-stressed and stressed plants (model’s precision: 0.91) as well as stress intensity (model’s precision: 0.84). These results highlighted the potential of machine learning techniques and automated image-based analysis in plant science. In the future, the models’ reliability and accuracy will be improved by adjusting the pipeline to different plant growth stages and specific leaf characteristics, and testing the model across different plant species and abiotic stresses (e.g. drought, temperature).