Polycystic ovarian syndrome which is commonly called as PCOS is a endocrine malfunction affecting women of reproductive age. Its diagnosis involves in detection of multiple small follicles mainly in the ovaries through ultrasound imaging. However, manual detection is time-consuming, subjective, and prone to errors. Hence, this study proposes an improvised follicle PCOS detection method using machine learning(Random Forest and Logistic Regression) from a sequence of given ultrasound images. The proposed method involves pre-processing the ultrasound images through IoT, followed by segmenting and extracting follicle features. Subsequently, a machine learning model is trained to classify the extracted features as normal or PCOS cases. The proposed method's performance is evaluated on a dataset of 400 ultrasound images from 50 patients, including 25 PCOS cases and 25 healthy controls. The experimental results demonstrate that the proposed method achieves a high classification accuracy of 93.75% and an AUC of 0.96. In addition, the proposed methodology outperforms in comparison with the state-of-the-art PCOS detection methods in terms of accuracy, sensitivity, specificity, and AUC. The proposed method also provides a quantitative measure of the severity of PCOS based on the number and size of the follicles detected.