Recently, a considerable amount of research has focused on understanding transportation mobility patterns from crowdsourced smartphone data. To this end, transportation mode detection is an indispensable, yet challenging task towards deriving meaningful information from large datasets collected using smartphones. Most studies to date use Global Navigation Satellite Systems (GNSS) derived data such as speed to detect transportation mode. Limited research relies on sensors that do not depend on external sources, such as accelerometer and gyroscope. The present work proposes a methodological framework based on machine learning for identifying the transportation mode using accelerometer, gyroscope and orientation data in the absence of battery consuming sensors, such as GNSS. Different models are developed and compared based on random forest and gradient boosting machine algorithms. A comparative study between GNSS free and GNSS based algorithms is also established. Results are further discussed and possible research directions are provided.