Talent development (TD) models typically stem from examining theoretical themes in isolation, and are predominantly analysed using statistical approaches best suited to experimental research. Whilst informative, these models offer a 'one size fits all' framework, and struggle to capture the holistic and dynamic nature of TD relevant to a specific expertise domain. This study aimed to holistically model TD in the sport of Olympic Weightlifting using Machine Learning. The development of 17 Junior weightlifting athletes (7 were identified as 'high performing' (i.e. talented) compared against the national average) was observed over a 12 month period. Each athlete was profiled on 385 theoretical driven attributes spanning a range of TD themes. These were then used to determine the optimal subset of attributes that best classified the high performing athletes from their low performing counterparts. A model of 15 attributes was identified as correctly classifying the groups with 100% accuracy. Encouragingly, this model is supportive of the separate themes that have emerged from past research that has investigated TD using 'traditional' statistical approaches where single (or relatively few) attributes are looked at in isolation. Our holistic model is discussed as encompassing Sampling and Specialising, Physiological factors, Psycho-social characteristics, Sibling effects, and Micro structure of practice as developmental themes. Importantly, whilst 385 theoretically driven attributes were input into the model, only 15 emerged as 'game changers' for 'early' expertise development. Findings support the need to observe TD holistically using appropriate methods, and for these observations to consider the domain specific antecedents of expertise development.