The technology space has seen a wide usage of Internet of things (IoT) in various domains such as smart homes, precision farming, and smart cities. This growing adoption has raised some concerns surrounding the security of IoT implementations, considering the demanding resource intensity for IoT devices. To address these security concerns, machine learning approaches have been proposed to mitigate attacks on IoT infrastructure. However, the heterogeneous nature and the large dimensionality of data generated by IoT devices proves to be challenging for machine learning approaches to accurately identify and mitigate attacks. Data dimensionality reduction has been used to reduce number of features in a dataset, by identifying and using only relevant features to reduce resource demands while maintaining accuracy. Feature selection is one of the methods used to achieve data dimensionality reduction. This paper compares three features selection techniques to see how their feature sets influence the accuracy of the classifier. Chi-square, recursive feature elimination, and select percentile techniques were used with the random forest classifier to select relevant features for the classifier. These techniques were considered to represent a diverse approach for feature selection methods. Chi-square and select percentile represent filter methods, while recursive feature elimination represent wrapper methods. The three chosen techniques are also amongst the most used techniques in their categories. The performance of the classifier was then evaluated based on accuracy, using the TON IoT dataset. The study shows that the random forest classifier was able to achieve an average accuracy of 99% with the feature set selected by the chi-square technique, followed by recursive feature elimination feature achieving 98%, while the feature sets selected by the select percentile technique achieved the lowest average accuracy score of 92%.