| 摘要: | 隨著時代的發展，我們的生活越來越便利，汽機車對於普遍大眾而言也已經是生活中一項不可或缺的交通工具。現今汽機車使用率逐年的增長，伴隨而來的交通事故也逐漸上升。由於道路交通事故層出不窮，不光是政府對於交通安全也提出許多的宣導政策及防範措施，社會大眾對於交通安全的議題也頗為重視。
近年來台灣的道路交通所造成的死亡人數雖有下降的趨勢，但在受傷人數以及肇事率方面仍然是持續在增加的。然而導致道路交通事故發生的原因有很多，例如氣候因素、人為因素、道路狀況等等，皆有可能是造成事故發生的重要關鍵。
本研究使用重複取樣技術來解決資料不平衡的問題，並使用各種不同特徵選取方式來找出和事故嚴重度有相關的各項因素，再使用決策樹、貝式分類、類神經網路以及最近鄰居法建立事故嚴重度預測模型，最後透過實驗，可以得知使用重複取樣技術在倍率為50倍時，並使用最近鄰居法所建立的預測模型效果是最好的。
As time goes by, our life is more and more convenient, automobiles and motorcy-cles have also become an indispensable means of transportation for the general public. Nowadays, the use of transportation has increased year by year, and the traffic accidents have also gradually increased. As a result of road traffic accidents, not only the government has advanced many policies and preventive measures, but the public attaches great importance to the issue of traffic safety
Although the number of deaths caused by road traffic in Taiwan has declined in recent years, it has continued to increase in terms of the number of injured and accidents. However, there are many causes of road traffic accidents, such as climatic factors, hu-man factors, road conditions, etc., all of which may be the key to the accident.
In this study, we used Synthetic Minority Over-sampling techniques to solve the problem of data imbalance, and used a variety of different feature selection methods to identify factors related to the severity of accidents, and then used decision tree, K-nearest neighbor, naïve bayesian and neural network for traffic accident severity pre-diction. Experimental results showed that using synthetic minority over-sampling tech-nique at a magnification of 50 and using the nearest neighbor method to establish the prediction model performed the best. |
