We present in this work a new mechanism based on an innovative hybrid approach for traffic and collision control in wireless ad hoc networks. We combine the well known equation based technique with the measurement based one in a way to obtain a predictive state of the network. Combining these two entries can also have correction effects on foretold values. These properties can be obtained by using the Kalman filter on the sender side. The sender carries out a loss probability prediction process, then performs a RED like algorithm before sending each packet. It may also adapt its sending rate depending on the network loss probability. We react to collisions since we show that losses due to collisions occur before congestion, therefore reacting to these losses can prevent reaching the congestion level. We have already developed a loss probability equation function of the input rate in the network. We also introduce a new loss probability measurement technique based on periodic feedbacks. Even though our formula may not calculate the exact loss probability value, due to some modeling simplifications, combining it to measurements will reduce and even remove these effects giving also a better prediction of future values.